From 3c718809d3d124e5156d240d660de15e7fd857fc Mon Sep 17 00:00:00 2001
From: Ilya Churaev <ilya.churaev@intel.com>
Date: Wed, 27 May 2020 18:37:19 +0300
Subject: [PATCH] Added ONNX reader for the OpenVINO (#532)

* Added ONNX reader for the OpenVINO

* Fixed comments

* Fixed comments

* Fixed message

* Fixed memory consumption

* Revert IReaderPtr

* Fixed Myriad tests

* Fixed comment

* Renamed inference_engine_ir_readers to inference_engine_ir_reader
---
 inference-engine/cmake/coverage_ie.cmake      |   2 +-
 inference-engine/cmake/plugins/plugins.cmake  |   4 +-
 .../src/openvino/inference_engine/ie_api.pyx  |  16 +-
 .../ie_bridges/python/tests/conftest.py       |  24 +-
 .../ie_bridges/python/tests/test_IECore.py    |  14 +-
 inference-engine/include/ie_api.h             | 109 +++---
 inference-engine/include/ie_plugin.hpp        |  22 --
 inference-engine/src/CMakeLists.txt           |   2 +-
 .../src/inference_engine/CMakeLists.txt       |   6 +-
 .../cnn_network_ngraph_impl.cpp               |   2 +
 .../src/inference_engine/ie_blob_stream.cpp   |  55 +++
 .../src/inference_engine/ie_core.cpp          | 181 ++++++----
 .../src/ir_readers/ie_ir_reader.cpp           |  85 -----
 .../src/ir_readers/ie_ir_reader.hpp           |  59 ----
 .../src/legacy_api/include/graph_tools.hpp    |   5 +-
 .../src/ie_cnn_layer_builder_ngraph.h         |   2 +-
 .../src/plugin_api/generic_ie.hpp             |   2 +
 inference-engine/src/readers/CMakeLists.txt   |  21 ++
 .../ir_reader}/CMakeLists.txt                 |   4 +-
 .../ir_reader}/ie_blob_proxy.hpp              |   0
 .../ir_reader}/ie_cnn_net_reader_impl.cpp     |  48 ++-
 .../ir_reader}/ie_cnn_net_reader_impl.h       |   6 +
 .../ir_reader}/ie_format_parser.cpp           |   0
 .../ir_reader}/ie_format_parser.h             |   0
 .../ir_reader}/ie_ir_parser.cpp               | 313 +++++++++++-------
 .../ir_reader}/ie_ir_parser.hpp               |  18 +-
 .../src/readers/ir_reader/ie_ir_reader.cpp    |  61 ++++
 .../src/readers/ir_reader/ie_ir_reader.hpp    |  72 ++++
 .../ir_reader}/ie_layer_parsers.cpp           |  15 +
 .../ir_reader}/ie_layer_parsers.h             |   0
 .../ir_reader}/parsers.h                      |   0
 .../src/readers/onnx_reader/CMakeLists.txt    |  41 +++
 .../readers/onnx_reader/ie_onnx_reader.cpp    |  36 ++
 .../readers/onnx_reader/ie_onnx_reader.hpp    |  48 +++
 .../src/readers/reader_api/ie_blob_stream.hpp |  35 ++
 .../src/readers/reader_api/ie_reader.hpp      |  63 ++++
 .../src/readers/reader_api/ie_reader_ptr.hpp  |  36 ++
 .../inference_engine/local_test.cpp           |   4 +-
 .../inference_engine/net_reader_test.cpp      |   2 +-
 .../onnx_reader/onnx_reader.cpp               | 136 ++++++++
 .../common_test_utils/test_assertions.hpp     |   6 +-
 .../functional_test_utils/network_utils.cpp   |   5 -
 .../unit_test_utils/CMakeLists.txt            |   4 +-
 .../behavior_test_plugin_infer_request.hpp    |   5 +-
 .../behavior_test_plugin_layers.hpp           |   2 +-
 .../plugin_tests/behavior_test_plugins.hpp    |   5 +-
 .../vpu/myriad_tests/vpu_watchdog_tests.cpp   |  14 +-
 .../network_tests/ngraph_network_test.cpp     |   8 +-
 .../tests_deprecated/helpers/CMakeLists.txt   |   2 +-
 .../tests_deprecated/unit/CMakeLists.txt      |   2 +-
 .../cnn_network/cnn_net_reader_impl_test.cpp  |   4 +-
 .../layers/internal/graph_concat_test.cpp     |   5 +-
 .../unit/graph_tools/graph_copy_tests.cpp     |   7 +-
 .../normalization/latest_in_fuse_test.cpp     |   9 +-
 54 files changed, 1146 insertions(+), 481 deletions(-)
 create mode 100644 inference-engine/src/inference_engine/ie_blob_stream.cpp
 delete mode 100644 inference-engine/src/ir_readers/ie_ir_reader.cpp
 delete mode 100644 inference-engine/src/ir_readers/ie_ir_reader.hpp
 create mode 100644 inference-engine/src/readers/CMakeLists.txt
 rename inference-engine/src/{ir_readers => readers/ir_reader}/CMakeLists.txt (89%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_blob_proxy.hpp (100%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_cnn_net_reader_impl.cpp (85%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_cnn_net_reader_impl.h (95%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_format_parser.cpp (100%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_format_parser.h (100%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_ir_parser.cpp (93%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_ir_parser.hpp (93%)
 create mode 100644 inference-engine/src/readers/ir_reader/ie_ir_reader.cpp
 create mode 100644 inference-engine/src/readers/ir_reader/ie_ir_reader.hpp
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_layer_parsers.cpp (94%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/ie_layer_parsers.h (100%)
 rename inference-engine/src/{ir_readers => readers/ir_reader}/parsers.h (100%)
 create mode 100644 inference-engine/src/readers/onnx_reader/CMakeLists.txt
 create mode 100644 inference-engine/src/readers/onnx_reader/ie_onnx_reader.cpp
 create mode 100644 inference-engine/src/readers/onnx_reader/ie_onnx_reader.hpp
 create mode 100644 inference-engine/src/readers/reader_api/ie_blob_stream.hpp
 create mode 100644 inference-engine/src/readers/reader_api/ie_reader.hpp
 create mode 100644 inference-engine/src/readers/reader_api/ie_reader_ptr.hpp
 create mode 100644 inference-engine/tests/functional/inference_engine/onnx_reader/onnx_reader.cpp

diff --git a/inference-engine/cmake/coverage_ie.cmake b/inference-engine/cmake/coverage_ie.cmake
index f288246bde2..67ec30bdaed 100644
--- a/inference-engine/cmake/coverage_ie.cmake
+++ b/inference-engine/cmake/coverage_ie.cmake
@@ -20,7 +20,7 @@ ie_coverage_genhtml(INFO_FILE "inference_engine"
                     PREFIX "${DLDT_COVERAGE_BASE_DIRECTORY}")
 
 ie_coverage_extract(INPUT "dldt" OUTPUT "inference_engine_ir_reader"
-                    PATTERNS "${DLDT_COVERAGE_BASE_DIRECTORY}/ir_readers/*")
+                    PATTERNS "${DLDT_COVERAGE_BASE_DIRECTORY}/readers/*")
 ie_coverage_genhtml(INFO_FILE "inference_engine_ir_reader"
                     PREFIX "${DLDT_COVERAGE_BASE_DIRECTORY}")
 
diff --git a/inference-engine/cmake/plugins/plugins.cmake b/inference-engine/cmake/plugins/plugins.cmake
index df874c6f040..e8907ae24e8 100644
--- a/inference-engine/cmake/plugins/plugins.cmake
+++ b/inference-engine/cmake/plugins/plugins.cmake
@@ -82,8 +82,8 @@ function(ie_add_plugin)
     if(TARGET inference_engine_preproc)
         add_dependencies(${IE_PLUGIN_NAME} inference_engine_preproc)
     endif()
-    if(TARGET inference_engine_ir_readers)
-        add_dependencies(${IE_PLUGIN_NAME} inference_engine_ir_readers)
+    if(TARGET inference_engine_ir_reader)
+        add_dependencies(${IE_PLUGIN_NAME} inference_engine_ir_reader)
     endif()
 
     # install rules
diff --git a/inference-engine/ie_bridges/python/src/openvino/inference_engine/ie_api.pyx b/inference-engine/ie_bridges/python/src/openvino/inference_engine/ie_api.pyx
index 06bf8b106a0..d406b747543 100644
--- a/inference-engine/ie_bridges/python/src/openvino/inference_engine/ie_api.pyx
+++ b/inference-engine/ie_bridges/python/src/openvino/inference_engine/ie_api.pyx
@@ -11,6 +11,7 @@ from libc.stdint cimport int64_t, uint8_t, int8_t, int32_t, uint16_t, int16_t
 from libc.string cimport memcpy
 
 import os
+from fnmatch import fnmatch
 from pathlib import Path
 import threading
 import warnings
@@ -268,20 +269,23 @@ cdef class IECore:
             net.impl = self.impl.readNetwork(xml_buffer, bin_buffer, len(weights))
             free(xml_buffer)
         else:
+            weights_ = "".encode()
             if isinstance(model, Path) and isinstance(weights, Path):
                 if not model.is_file():
                     raise Exception("Path to the model {} doesn't exist or it's a directory".format(model))
-                if not weights.is_file():
-                    raise Exception("Path to the weights {} doesn't exist or it's a directory".format(weights))
+                if model.suffix != ".onnx":
+                    if not weights.is_file():
+                        raise Exception("Path to the weights {} doesn't exist or it's a directory".format(weights))
+                    weights_ = bytes(weights)
                 model_ = bytes(model)
-                weights_ = bytes(weights)
             else:
                 if not os.path.isfile(model):
                     raise Exception("Path to the model {} doesn't exist or it's a directory".format(model))
-                if not os.path.isfile(weights):
-                    raise Exception("Path to the weights {} doesn't exist or it's a directory".format(weights))
+                if not fnmatch(model, "*.onnx"):
+                    if not os.path.isfile(weights):
+                        raise Exception("Path to the weights {} doesn't exist or it's a directory".format(weights))
+                    weights_ = weights.encode()
                 model_ = model.encode()
-                weights_ = weights.encode()
             net.impl =  self.impl.readNetwork(model_, weights_)
         return net
 
diff --git a/inference-engine/ie_bridges/python/tests/conftest.py b/inference-engine/ie_bridges/python/tests/conftest.py
index 81910f911af..ae86c64beb6 100644
--- a/inference-engine/ie_bridges/python/tests/conftest.py
+++ b/inference-engine/ie_bridges/python/tests/conftest.py
@@ -3,10 +3,7 @@ import pytest
 
 
 def model_path(is_myriad=False):
-    if os.environ.get("MODELS_PATH"):
-        path_to_repo = os.environ.get("MODELS_PATH")
-    else:
-        raise EnvironmentError("MODELS_PATH variable isn't set")
+    path_to_repo = os.environ["MODELS_PATH"]
     if not is_myriad:
         test_xml = os.path.join(path_to_repo, "models", "test_model", 'test_model_fp32.xml')
         test_bin = os.path.join(path_to_repo, "models", "test_model", 'test_model_fp32.bin')
@@ -15,24 +12,27 @@ def model_path(is_myriad=False):
         test_bin = os.path.join(path_to_repo, "models", "test_model", 'test_model_fp16.bin')
     return (test_xml, test_bin)
 
+
+def model_onnx_path():
+    path_to_repo = os.environ["MODELS_PATH"]
+    test_onnx = os.path.join(path_to_repo, "models", "test_model", 'test_model.onnx')
+    return test_onnx
+
+
 def image_path():
-    if os.environ.get("DATA_PATH"):
-        path_to_repo = os.environ.get("DATA_PATH")
-    else:
-        raise EnvironmentError("DATA_PATH variable isn't set")
+    path_to_repo = os.environ["DATA_PATH"]
     path_to_img = os.path.join(path_to_repo, 'validation_set', '224x224', 'dog.bmp')
     return path_to_img
 
+
 def plugins_path():
-    if os.environ.get("DATA_PATH"):
-        path_to_repo = os.environ.get("DATA_PATH")
-    else:
-        raise EnvironmentError("DATA_PATH variable isn't set")
+    path_to_repo = os.environ["DATA_PATH"]
     plugins_xml = os.path.join(path_to_repo, 'ie_class', 'plugins.xml')
     plugins_win_xml = os.path.join(path_to_repo, 'ie_class', 'plugins_mingw.xml')
     plugins_osx_xml = os.path.join(path_to_repo, 'ie_class', 'plugins_apple.xml')
     return (plugins_xml, plugins_win_xml, plugins_osx_xml)
 
+
 @pytest.fixture(scope='session')
 def device():
     return os.environ.get("TEST_DEVICE") if os.environ.get("TEST_DEVICE") else "CPU"
diff --git a/inference-engine/ie_bridges/python/tests/test_IECore.py b/inference-engine/ie_bridges/python/tests/test_IECore.py
index 9eabe34ccc6..ac97e200c0a 100644
--- a/inference-engine/ie_bridges/python/tests/test_IECore.py
+++ b/inference-engine/ie_bridges/python/tests/test_IECore.py
@@ -5,7 +5,7 @@ import numpy as np
 from pathlib import Path
 
 from openvino.inference_engine import IENetwork, IECore, ExecutableNetwork
-from conftest import model_path, plugins_path
+from conftest import model_path, plugins_path, model_onnx_path
 
 
 test_net_xml, test_net_bin = model_path()
@@ -143,28 +143,39 @@ def test_get_metric_str():
     assert isinstance(param, str), "Parameter value for 'FULL_DEVICE_NAME' " \
                                    "metric must be string but {} is returned".format(type(param))
 
+
 def test_read_network_from_xml():
     ie = IECore()
     net = ie.read_network(model=test_net_xml, weights=test_net_bin)
     assert isinstance(net, IENetwork)
 
+
 def test_read_network_as_path():
     ie = IECore()
     net = ie.read_network(model=Path(model_path()[0]), weights=Path(test_net_bin))
     assert isinstance(net, IENetwork)
 
+
+def test_read_network_from_onnx():
+    ie = IECore()
+    net = ie.read_network(model=model_onnx_path())
+    assert isinstance(net, IENetwork)
+
+
 def test_incorrect_xml():
     ie = IECore()
     with pytest.raises(Exception) as e:
         ie.read_network(model="./model.xml", weights=Path(test_net_bin))
     assert "Path to the model ./model.xml doesn't exist or it's a directory" in str(e.value)
 
+
 def test_incorrect_bin():
     ie = IECore()
     with pytest.raises(Exception) as e:
         ie.read_network(model=test_net_xml, weights="./model.bin")
     assert "Path to the weights ./model.bin doesn't exist or it's a directory" in str(e.value)
 
+
 def test_read_net_from_buffer():
     ie = IECore()
     with open(test_net_bin, 'rb') as f:
@@ -174,6 +185,7 @@ def test_read_net_from_buffer():
     net = ie.read_network(model=xml, weights=bin, init_from_buffer=True)
     assert isinstance(net, IENetwork)
 
+
 def test_net_from_buffer_valid():
     ie = IECore()
     with open(test_net_bin, 'rb') as f:
diff --git a/inference-engine/include/ie_api.h b/inference-engine/include/ie_api.h
index a16e0eae6d7..5faf7bd40de 100644
--- a/inference-engine/include/ie_api.h
+++ b/inference-engine/include/ie_api.h
@@ -12,39 +12,38 @@
 #include "details/ie_no_copy.hpp"
 
 #if defined(USE_STATIC_IE) || (defined(__GNUC__) && (__GNUC__ < 4))
-#define INFERENCE_ENGINE_API(...) extern "C" __VA_ARGS__
-#define INFERENCE_ENGINE_API_CPP(...) __VA_ARGS__
-#define INFERENCE_ENGINE_API_CLASS(...) __VA_ARGS__
-#define INFERENCE_ENGINE_CDECL __attribute__((cdecl))
+# define INFERENCE_ENGINE_API(...) extern "C" __VA_ARGS__
+# define INFERENCE_ENGINE_API_CPP(...) __VA_ARGS__
+# define INFERENCE_ENGINE_API_CLASS(...) __VA_ARGS__
+# define INFERENCE_ENGINE_CDECL __attribute__((cdecl))
 #else
-#if defined(_WIN32)
-#define INFERENCE_ENGINE_CDECL
-
-#ifdef IMPLEMENT_INFERENCE_ENGINE_API
-#define INFERENCE_ENGINE_API(...) extern "C" __declspec(dllexport) __VA_ARGS__ __cdecl
-#define INFERENCE_ENGINE_API_CPP(...) __declspec(dllexport) __VA_ARGS__ __cdecl
-#define INFERENCE_ENGINE_API_CLASS(...) __declspec(dllexport) __VA_ARGS__
-#else
-#define INFERENCE_ENGINE_API(...) extern "C" __declspec(dllimport) __VA_ARGS__ __cdecl
-#define INFERENCE_ENGINE_API_CPP(...) __declspec(dllimport) __VA_ARGS__ __cdecl
-#define INFERENCE_ENGINE_API_CLASS(...) __declspec(dllimport) __VA_ARGS__
-#endif
-#else
-#define INFERENCE_ENGINE_CDECL __attribute__((cdecl))
-#define INFERENCE_ENGINE_API(...) extern "C" __attribute__((visibility("default"))) __VA_ARGS__
-#define INFERENCE_ENGINE_API_CPP(...) __attribute__((visibility("default"))) __VA_ARGS__
-#define INFERENCE_ENGINE_API_CLASS(...) __attribute__((visibility("default"))) __VA_ARGS__
-#endif
+# if defined(_WIN32)
+#  define INFERENCE_ENGINE_CDECL
+#  ifdef IMPLEMENT_INFERENCE_ENGINE_API
+#   define INFERENCE_ENGINE_API(...) extern "C" __declspec(dllexport) __VA_ARGS__ __cdecl
+#   define INFERENCE_ENGINE_API_CPP(...) __declspec(dllexport) __VA_ARGS__ __cdecl
+#   define INFERENCE_ENGINE_API_CLASS(...) __declspec(dllexport) __VA_ARGS__
+#  else
+#   define INFERENCE_ENGINE_API(...) extern "C" __declspec(dllimport) __VA_ARGS__ __cdecl
+#   define INFERENCE_ENGINE_API_CPP(...) __declspec(dllimport) __VA_ARGS__ __cdecl
+#   define INFERENCE_ENGINE_API_CLASS(...) __declspec(dllimport) __VA_ARGS__
+#  endif
+# else
+#  define INFERENCE_ENGINE_CDECL __attribute__((cdecl))
+#  define INFERENCE_ENGINE_API(...) extern "C" __attribute__((visibility("default"))) __VA_ARGS__
+#  define INFERENCE_ENGINE_API_CPP(...) __attribute__((visibility("default"))) __VA_ARGS__
+#  define INFERENCE_ENGINE_API_CLASS(...) __attribute__((visibility("default"))) __VA_ARGS__
+# endif
 #endif
 
 #if defined(_WIN32)
-#define INFERENCE_ENGINE_DEPRECATED(msg) __declspec(deprecated(msg))
+# define INFERENCE_ENGINE_DEPRECATED(msg) __declspec(deprecated(msg))
 #elif defined __INTEL_COMPILER
-#define INFERENCE_ENGINE_DEPRECATED(msg) __attribute__((deprecated(msg)))
+# define INFERENCE_ENGINE_DEPRECATED(msg) __attribute__((deprecated(msg)))
 #elif defined(__GNUC__)
-#define INFERENCE_ENGINE_DEPRECATED(msg) __attribute__((deprecated((msg))))
+# define INFERENCE_ENGINE_DEPRECATED(msg) __attribute__((deprecated((msg))))
 #else
-#define INFERENCE_ENGINE_DEPRECATED(msg)
+# define INFERENCE_ENGINE_DEPRECATED(msg)
 #endif
 
 #if defined IMPLEMENT_INFERENCE_ENGINE_API || defined IMPLEMENT_INFERENCE_ENGINE_PLUGIN
@@ -63,32 +62,32 @@
 
 // Suppress warning "-Wdeprecated-declarations" / C4996
 #if defined(_MSC_VER)
-#define IE_DO_PRAGMA(x) __pragma(x)
+# define IE_DO_PRAGMA(x) __pragma(x)
 #elif defined(__GNUC__)
-#define IE_DO_PRAGMA(x) _Pragma(#x)
+# define IE_DO_PRAGMA(x) _Pragma(#x)
 #else
-#define IE_DO_PRAGMA(x)
+# define IE_DO_PRAGMA(x)
 #endif
 
 #if defined(_MSC_VER) && !defined(__clang__)
-#define IE_SUPPRESS_DEPRECATED_START \
-    IE_DO_PRAGMA(warning(push))      \
-    IE_DO_PRAGMA(warning(disable : 4996))
-#define IE_SUPPRESS_DEPRECATED_END IE_DO_PRAGMA(warning(pop))
+# define IE_SUPPRESS_DEPRECATED_START \
+     IE_DO_PRAGMA(warning(push))      \
+     IE_DO_PRAGMA(warning(disable : 4996))
+# define IE_SUPPRESS_DEPRECATED_END IE_DO_PRAGMA(warning(pop))
 #elif defined(__INTEL_COMPILER)
-#define IE_SUPPRESS_DEPRECATED_START \
-    IE_DO_PRAGMA(warning(push))      \
-    IE_DO_PRAGMA(warning(disable : 1478))
-    IE_DO_PRAGMA(warning(disable : 1786))
-#define IE_SUPPRESS_DEPRECATED_END IE_DO_PRAGMA(warning(pop))
+# define IE_SUPPRESS_DEPRECATED_START \
+     IE_DO_PRAGMA(warning(push))      \
+     IE_DO_PRAGMA(warning(disable : 1478))
+     IE_DO_PRAGMA(warning(disable : 1786))
+# define IE_SUPPRESS_DEPRECATED_END IE_DO_PRAGMA(warning(pop))
 #elif defined(__clang__) || ((__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ > 405))
-#define IE_SUPPRESS_DEPRECATED_START  \
-    IE_DO_PRAGMA(GCC diagnostic push) \
-    IE_DO_PRAGMA(GCC diagnostic ignored "-Wdeprecated-declarations")
-#define IE_SUPPRESS_DEPRECATED_END IE_DO_PRAGMA(GCC diagnostic pop)
+# define IE_SUPPRESS_DEPRECATED_START  \
+     IE_DO_PRAGMA(GCC diagnostic push) \
+     IE_DO_PRAGMA(GCC diagnostic ignored "-Wdeprecated-declarations")
+# define IE_SUPPRESS_DEPRECATED_END IE_DO_PRAGMA(GCC diagnostic pop)
 #else
-#define IE_SUPPRESS_DEPRECATED_START
-#define IE_SUPPRESS_DEPRECATED_END
+# define IE_SUPPRESS_DEPRECATED_START
+# define IE_SUPPRESS_DEPRECATED_END
 #endif
 
 #ifdef _WIN32
@@ -111,3 +110,25 @@
 #  define ENABLE_UNICODE_PATH_SUPPORT
 # endif
 #endif
+
+/**
+ * @def INFERENCE_PLUGIN_API(type)
+ * @brief Defines Inference Engine Plugin API method
+ * @param type A plugin type
+ */
+
+#if defined(_WIN32)
+# ifdef IMPLEMENT_INFERENCE_ENGINE_PLUGIN
+#  define INFERENCE_PLUGIN_API(type) extern "C" __declspec(dllexport) type
+# else
+#  define INFERENCE_PLUGIN_API(type) extern "C" type
+# endif
+#elif (__GNUC__ >= 4)  // NOLINT
+# ifdef IMPLEMENT_INFERENCE_ENGINE_PLUGIN
+#  define INFERENCE_PLUGIN_API(type) extern "C" __attribute__((visibility("default"))) type
+# else
+#  define INFERENCE_PLUGIN_API(type) extern "C" type
+# endif
+#else
+# define INFERENCE_PLUGIN_API(TYPE) extern "C" TYPE
+#endif
diff --git a/inference-engine/include/ie_plugin.hpp b/inference-engine/include/ie_plugin.hpp
index 82b975cd6b9..26e59824465 100644
--- a/inference-engine/include/ie_plugin.hpp
+++ b/inference-engine/include/ie_plugin.hpp
@@ -25,28 +25,6 @@
 #include "ie_iexecutable_network.hpp"
 #include "ie_version.hpp"
 
-/**
- * @def INFERENCE_PLUGIN_API(type)
- * @brief Defines Inference Engine Plugin API method
- * @param type A plugin type
- */
-
-#if defined(_WIN32)
-#ifdef IMPLEMENT_INFERENCE_ENGINE_PLUGIN
-#define INFERENCE_PLUGIN_API(type) extern "C" __declspec(dllexport) type
-#else
-#define INFERENCE_PLUGIN_API(type) extern "C" type
-#endif
-#elif (__GNUC__ >= 4)  // NOLINT
-#ifdef IMPLEMENT_INFERENCE_ENGINE_PLUGIN
-#define INFERENCE_PLUGIN_API(type) extern "C" __attribute__((visibility("default"))) type
-#else
-#define INFERENCE_PLUGIN_API(type) extern "C" type
-#endif
-#else
-#define INFERENCE_PLUGIN_API(TYPE) extern "C" TYPE
-#endif
-
 namespace InferenceEngine {
 
 /**
diff --git a/inference-engine/src/CMakeLists.txt b/inference-engine/src/CMakeLists.txt
index a97e1c780ee..0ba0dd7849e 100644
--- a/inference-engine/src/CMakeLists.txt
+++ b/inference-engine/src/CMakeLists.txt
@@ -4,7 +4,7 @@
 
 add_subdirectory(preprocessing)
 
-add_subdirectory(ir_readers)
+add_subdirectory(readers)
 
 add_subdirectory(legacy_api)
 
diff --git a/inference-engine/src/inference_engine/CMakeLists.txt b/inference-engine/src/inference_engine/CMakeLists.txt
index 091e68f1972..ce94bb3567f 100644
--- a/inference-engine/src/inference_engine/CMakeLists.txt
+++ b/inference-engine/src/inference_engine/CMakeLists.txt
@@ -123,7 +123,7 @@ target_include_directories(${TARGET_NAME}_obj SYSTEM PRIVATE $<TARGET_PROPERTY:n
                                                              $<TARGET_PROPERTY:pugixml,INTERFACE_INCLUDE_DIRECTORIES>)
 
 target_include_directories(${TARGET_NAME}_obj PRIVATE "${CMAKE_CURRENT_SOURCE_DIR}"
-                                                      $<TARGET_PROPERTY:${TARGET_NAME}_ir_readers,INTERFACE_INCLUDE_DIRECTORIES>
+                                                      $<TARGET_PROPERTY:${TARGET_NAME}_reader_api,INTERFACE_INCLUDE_DIRECTORIES>
                                                       $<TARGET_PROPERTY:${TARGET_NAME}_plugin_api,INTERFACE_INCLUDE_DIRECTORIES>)
 
 if(ENABLE_PROFILING_ITT AND INTEL_ITT_LIBS)
@@ -158,8 +158,8 @@ target_link_libraries(${TARGET_NAME} PRIVATE Threads::Threads ${INTEL_ITT_LIBS}
                                              ${NGRAPH_LIBRARIES} inference_engine_transformations
                                      PUBLIC ${CMAKE_DL_LIBS} ${TARGET_NAME}_legacy)
 
-target_include_directories(${TARGET_NAME} INTERFACE ${PUBLIC_HEADERS_DIR}
-                                          PRIVATE $<TARGET_PROPERTY:${TARGET_NAME}_plugin_api,INTERFACE_INCLUDE_DIRECTORIES>)
+target_include_directories(${TARGET_NAME} INTERFACE ${PUBLIC_HEADERS_DIR} PRIVATE
+                                                $<TARGET_PROPERTY:${TARGET_NAME}_plugin_api,INTERFACE_INCLUDE_DIRECTORIES>)
 
 if(WIN32)
     set_target_properties(${TARGET_NAME} PROPERTIES COMPILE_PDB_NAME ${TARGET_NAME})
diff --git a/inference-engine/src/inference_engine/cnn_network_ngraph_impl.cpp b/inference-engine/src/inference_engine/cnn_network_ngraph_impl.cpp
index 16190983511..6e2dfcb80bf 100644
--- a/inference-engine/src/inference_engine/cnn_network_ngraph_impl.cpp
+++ b/inference-engine/src/inference_engine/cnn_network_ngraph_impl.cpp
@@ -82,6 +82,8 @@ CNNNetwork::CNNNetwork(const std::shared_ptr<const ngraph::Function>& graph) {
     }
 
     // Copy nGraph function
+    if (graph == nullptr)
+        THROW_IE_EXCEPTION << "Cannot create CNNNetwork from empty nGraph function!";
     network = std::make_shared<CNNNetworkNGraphImpl>(copyFunction(graph, false, {}));
     actual = network.get();
     if (actual == nullptr) {
diff --git a/inference-engine/src/inference_engine/ie_blob_stream.cpp b/inference-engine/src/inference_engine/ie_blob_stream.cpp
new file mode 100644
index 00000000000..24cb3680fc7
--- /dev/null
+++ b/inference-engine/src/inference_engine/ie_blob_stream.cpp
@@ -0,0 +1,55 @@
+// Copyright (C) 2018-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "ie_blob_stream.hpp"
+
+#include <ie_blob.h>
+#include <istream>
+
+InferenceEngine::details::BlobStream::BlobBuffer::BlobBuffer(const InferenceEngine::Blob::CPtr& blob) {
+    char* data = nullptr;
+    std::streampos size;
+    if (!blob) {
+        size = 0;
+    } else {
+        data = blob->cbuffer().as<char*>();
+        size = blob->byteSize();
+    }
+    setg(data, data, data + size);
+}
+InferenceEngine::details::BlobStream::BlobBuffer::~BlobBuffer() {}
+
+std::streampos InferenceEngine::details::BlobStream::BlobBuffer::seekpos(std::streampos sp, std::ios_base::openmode which) {
+    if (!(which & ios_base::in))
+        return streampos(-1);
+    if (sp < 0 || sp > egptr() - eback())
+        return streampos(-1);
+    setg(eback(), eback() + sp, egptr());
+    return sp;
+}
+std::streampos InferenceEngine::details::BlobStream::BlobBuffer::seekoff(std::streamoff off, std::ios_base::seekdir way, std::ios_base::openmode which) {
+    if (!(which & std::ios_base::in))
+        return streampos(-1);
+    switch (way) {
+    default:
+    case std::ios_base::beg:
+        setg(eback(), eback() + off, egptr());
+        break;
+    case std::ios_base::cur:
+        gbump(off);
+        break;
+    case std::ios_base::end:
+        setg(eback(), egptr() + off, egptr());
+        break;
+    }
+    return gptr() - eback();
+}
+
+InferenceEngine::Blob::CPtr InferenceEngine::details::BlobStream::getBlob() {
+    return blob;
+}
+
+InferenceEngine::details::BlobStream::BlobStream(const InferenceEngine::Blob::CPtr& blob): buffer(blob), std::ios(0), std::istream(&buffer), blob(blob) {}
+
+InferenceEngine::details::BlobStream::~BlobStream() {}
diff --git a/inference-engine/src/inference_engine/ie_core.cpp b/inference-engine/src/inference_engine/ie_core.cpp
index ff71ea1a45d..0aab9219911 100644
--- a/inference-engine/src/inference_engine/ie_core.cpp
+++ b/inference-engine/src/inference_engine/ie_core.cpp
@@ -11,11 +11,15 @@
 #include <map>
 #include <memory>
 #include <sstream>
+#include <streambuf>
 #include <string>
 #include <utility>
 #include <vector>
+#include <istream>
 #include <mutex>
 
+#include "ie_blob_stream.hpp"
+#include <ie_reader_ptr.hpp>
 #include <ngraph/opsets/opset.hpp>
 #include "cpp/ie_cnn_net_reader.h"
 #include "cpp/ie_plugin_cpp.hpp"
@@ -44,7 +48,7 @@ InferenceEngine::details::SharedObjectLoader::Ptr cnnReaderLoader;
 
 InferenceEngine::details::SharedObjectLoader::Ptr createCnnReaderLoader() {
     std::call_once(flag, [&] () {
-        FileUtils::FilePath libraryName = FileUtils::toFilePath(std::string("inference_engine_ir_readers") + std::string(IE_BUILD_POSTFIX));
+        FileUtils::FilePath libraryName = FileUtils::toFilePath(std::string("inference_engine_ir_reader") + std::string(IE_BUILD_POSTFIX));
         FileUtils::FilePath irReadersLibraryPath = FileUtils::makeSharedLibraryName(getInferenceEngineLibraryPath(), libraryName);
 
         if (!FileUtils::fileExist(irReadersLibraryPath)) {
@@ -129,6 +133,79 @@ Parameter copyParameterValue(const Parameter & value) {
 
 }  // namespace
 
+class Reader: public IReader {
+private:
+    InferenceEngine::IReaderPtr ptr;
+    std::once_flag readFlag;
+    std::string name;
+    std::string location;
+
+    InferenceEngine::IReaderPtr getReaderPtr() {
+        std::call_once(readFlag, [&] () {
+            FileUtils::FilePath libraryName = FileUtils::toFilePath(location);
+            FileUtils::FilePath readersLibraryPath = FileUtils::makeSharedLibraryName(getInferenceEngineLibraryPath(), libraryName);
+
+            if (!FileUtils::fileExist(readersLibraryPath)) {
+                THROW_IE_EXCEPTION << "Please, make sure that Inference Engine ONNX reader library "
+                    << FileUtils::fromFilePath(::FileUtils::makeSharedLibraryName({}, libraryName)) << " is in "
+                    << getIELibraryPath();
+            }
+            ptr = IReaderPtr(readersLibraryPath);
+        });
+
+        return ptr;
+    }
+
+    InferenceEngine::IReaderPtr getReaderPtr() const {
+        return const_cast<Reader*>(this)->getReaderPtr();
+    }
+
+    void Release() noexcept override {
+        delete this;
+    }
+
+public:
+    using Ptr = std::shared_ptr<Reader>;
+    Reader(const std::string& name, const std::string location): name(name), location(location) {}
+    bool supportModel(std::istream& model) const override {
+        auto reader = getReaderPtr();
+        return reader->supportModel(model);
+    }
+    CNNNetwork read(std::istream& model, const std::vector<IExtensionPtr>& exts) const override {
+        auto reader = getReaderPtr();
+        return reader->read(model, exts);
+    }
+    CNNNetwork read(std::istream& model, std::istream& weights, const std::vector<IExtensionPtr>& exts) const override {
+        auto reader = getReaderPtr();
+        return reader->read(model, weights, exts);
+    }
+    std::vector<std::string> getDataFileExtensions() const override {
+        auto reader = getReaderPtr();
+        return reader->getDataFileExtensions();
+    }
+    std::string getName() const {
+        return name;
+    }
+};
+
+namespace {
+
+// Extension to plugins creator
+std::multimap<std::string, Reader::Ptr> readers;
+
+void registerReaders() {
+    static std::mutex readerMutex;
+    std::lock_guard<std::mutex> lock(readerMutex);
+    // TODO: Read readers info from XML
+    auto onnxReader = std::make_shared<Reader>("ONNX", std::string("inference_engine_onnx_reader") + std::string(IE_BUILD_POSTFIX));
+    readers.emplace("onnx", onnxReader);
+    readers.emplace("prototxt", onnxReader);
+    auto irReader = std::make_shared<Reader>("IR", std::string("inference_engine_ir_reader") + std::string(IE_BUILD_POSTFIX));
+    readers.emplace("xml", irReader);
+}
+
+}  // namespace
+
 CNNNetReaderPtr CreateCNNNetReaderPtr() noexcept {
     auto loader = createCnnReaderLoader();
     return CNNNetReaderPtr(loader);
@@ -207,20 +284,6 @@ class Core::Impl : public ICore {
         std::vector<FileUtils::FilePath> listOfExtentions;
     };
 
-    /**
-     * @brief Holds original blob in order to avoid situations
-     *        when original blob is allocated on stack
-     */
-    class WeightsHolderBlob : public TBlob<uint8_t> {
-        Blob::CPtr originBlob;
-
-    public:
-        explicit WeightsHolderBlob(const Blob::CPtr& weights) :
-            TBlob<uint8_t>(weights->getTensorDesc(),
-                           weights->cbuffer().as<uint8_t*>()),
-            originBlob(weights) { }
-    };
-
     std::unordered_set<std::string> opsetNames;
     std::vector<IExtensionPtr> extensions;
 
@@ -311,58 +374,57 @@ public:
 
     CNNNetwork ReadNetwork(const std::string& modelPath, const std::string& binPath) const override {
         IE_PROFILING_AUTO_SCOPE(Core::ReadNetwork)
-        IE_SUPPRESS_DEPRECATED_START
-        ResponseDesc desc;
-        CNNNetReaderPtr cnnReader(createCnnReaderLoader());
-        StatusCode rt = cnnReader->ReadNetwork(modelPath.c_str(), &desc);
-        if (rt != OK) THROW_IE_EXCEPTION << desc.msg;
-        if (cnnReader->getVersion(&desc) >= 10) {
-            std::lock_guard<std::mutex> lock(pluginsMutex);
-            cnnReader->addExtensions(GetExtensions());
-        }
-        std::string bPath = binPath;
-        if (bPath.empty()) {
-            bPath = modelPath;
-            auto pos = bPath.rfind('.');
-            if (pos != std::string::npos) bPath = bPath.substr(0, pos);
-            bPath += ".bin";
 
-            if (!FileUtils::fileExist(bPath)) bPath.clear();
-        }
+        std::ifstream modelStream(modelPath, std::ios::binary);
+        if (!modelStream.is_open())
+            THROW_IE_EXCEPTION << "Model file " << modelPath << " cannot be opened!";
 
-        if (!bPath.empty()) {
-            rt = cnnReader->ReadWeights(bPath.c_str(), &desc);
-            if (rt != OK) THROW_IE_EXCEPTION << desc.msg;
-        } else {
-            TBlob<uint8_t>::Ptr weights_ptr;
-            rt = cnnReader->SetWeights(weights_ptr, &desc);
-            if (rt != OK) THROW_IE_EXCEPTION << desc.msg;
+        auto fileExt = modelPath.substr(modelPath.find_last_of(".") + 1);
+        for (auto it = readers.lower_bound(fileExt); it != readers.upper_bound(fileExt); it++) {
+            auto reader = it->second;
+            if (reader->supportModel(modelStream)) {
+                // Find weights
+                std::string bPath = binPath;
+                if (bPath.empty()) {
+                    auto pathWoExt = modelPath;
+                    auto pos = modelPath.rfind('.');
+                    if (pos != std::string::npos) pathWoExt = modelPath.substr(0, pos);
+                    for (const auto& ext : reader->getDataFileExtensions()) {
+                        bPath = pathWoExt + "." + ext;
+                        if (!FileUtils::fileExist(bPath)) {
+                            bPath.clear();
+                        } else {
+                            break;
+                        }
+                    }
+                }
+                if (!bPath.empty()) {
+                    std::ifstream binStream;
+                    binStream.open(bPath, std::ios::binary);
+                    if (!binStream.is_open())
+                        THROW_IE_EXCEPTION << "Weights file " << bPath << " cannot be opened!";
+                    return reader->read(modelStream, binStream, extensions);
+                }
+                return reader->read(modelStream, extensions);
+            }
         }
-        IE_SUPPRESS_DEPRECATED_END
-
-        return CNNNetwork(cnnReader);
+        THROW_IE_EXCEPTION << "Unknown model format! Cannot read the model: " << modelPath;
     }
 
     CNNNetwork ReadNetwork(const std::string& model, const Blob::CPtr& weights) const override {
         IE_PROFILING_AUTO_SCOPE(Core::ReadNetwork)
-        IE_SUPPRESS_DEPRECATED_START
-        ResponseDesc desc;
-        CNNNetReaderPtr cnnReader(createCnnReaderLoader());
-        StatusCode rt = cnnReader->ReadNetwork(model.data(), model.length(), &desc);
-        if (rt != OK) THROW_IE_EXCEPTION << desc.msg;
-        if (cnnReader->getVersion(&desc) >= 10) {
-            std::lock_guard<std::mutex> lock(pluginsMutex);
-            cnnReader->addExtensions(GetExtensions());
-        }
-        TBlob<uint8_t>::Ptr weights_ptr;
-        if (weights) {
-            weights_ptr = std::make_shared<WeightsHolderBlob>(weights);
-        }
-        rt = cnnReader->SetWeights(weights_ptr, &desc);
-        if (rt != OK) THROW_IE_EXCEPTION << desc.msg;
-        IE_SUPPRESS_DEPRECATED_END
+        std::istringstream modelStream(model);
+        details::BlobStream binStream(weights);
 
-        return CNNNetwork(cnnReader);
+        for (auto it = readers.begin(); it != readers.end(); it++) {
+            auto reader = it->second;
+            if (reader->supportModel(modelStream)) {
+                if (weights)
+                    return reader->read(modelStream, binStream, extensions);
+                return reader->read(modelStream, extensions);
+            }
+        }
+        THROW_IE_EXCEPTION << "Unknown model format! Cannot read the model from string!";
     }
 
     ExecutableNetwork LoadNetwork(const CNNNetwork& network, const std::string& deviceName,
@@ -642,6 +704,7 @@ Core::Impl::Impl() {
     opsetNames.insert("opset1");
     opsetNames.insert("opset2");
     opsetNames.insert("opset3");
+    registerReaders();
 }
 
 Core::Impl::~Impl() {}
diff --git a/inference-engine/src/ir_readers/ie_ir_reader.cpp b/inference-engine/src/ir_readers/ie_ir_reader.cpp
deleted file mode 100644
index ec6badfdcef..00000000000
--- a/inference-engine/src/ir_readers/ie_ir_reader.cpp
+++ /dev/null
@@ -1,85 +0,0 @@
-// Copyright (C) 2018-2020 Intel Corporation
-// SPDX-License-Identifier: Apache-2.0
-//
-
-#include <file_utils.h>
-#include <xml_parse_utils.h>
-
-#include <ie_ir_reader.hpp>
-#include <memory>
-#include <ngraph/ngraph.hpp>
-#include <string>
-#include <vector>
-
-#include "description_buffer.hpp"
-#include "ie_ir_parser.hpp"
-#include "ie_ngraph_utils.hpp"
-
-using namespace InferenceEngine;
-
-static size_t GetIRVersion(pugi::xml_node& root) {
-    return XMLParseUtils::GetUIntAttr(root, "version", 0);
-}
-
-std::shared_ptr<ngraph::Function> IRReader::read(const std::string& modelPath, const std::string& binPath) {
-    std::ifstream modelFile(modelPath);
-    if (!modelFile.is_open()) THROW_IE_EXCEPTION << "File " << modelPath << " cannot be openned!";
-
-    std::stringstream modelBuf;
-    modelBuf << modelFile.rdbuf();
-
-    Blob::Ptr weights;
-    std::string bPath = binPath;
-    if (bPath.empty()) {
-        bPath = modelPath;
-        auto pos = bPath.rfind('.');
-        if (pos != std::string::npos) bPath = bPath.substr(0, pos);
-        bPath += ".bin";
-
-        if (!FileUtils::fileExist(bPath)) bPath.clear();
-    }
-
-    if (!bPath.empty()) {
-        int64_t fileSize = FileUtils::fileSize(bPath);
-
-        if (fileSize < 0)
-            THROW_IE_EXCEPTION << "Filesize for: " << bPath << " - " << fileSize
-                               << " < 0. Please, check weights file existence.";
-
-        size_t ulFileSize = static_cast<size_t>(fileSize);
-
-        weights = make_shared_blob<uint8_t>(TensorDesc(Precision::U8, {ulFileSize}, Layout::C));
-        weights->allocate();
-        FileUtils::readAllFile(bPath, weights->buffer(), ulFileSize);
-    }
-
-    return read(modelBuf.str(), weights);
-}
-
-std::shared_ptr<ngraph::Function> IRReader::read(const std::string& model, const Blob::CPtr& weights) {
-    pugi::xml_document xmlDoc;
-    pugi::xml_parse_result res = xmlDoc.load_buffer(model.data(), model.length());
-    if (res.status != pugi::status_ok) {
-        THROW_IE_EXCEPTION << res.description() << "at offset " << res.offset;
-    }
-    return readXml(xmlDoc, weights);
-}
-
-std::shared_ptr<ngraph::Function> IRReader::readXml(const pugi::xml_document& xmlDoc, const Blob::CPtr& weights) {
-    try {
-        // check which version it is...
-        pugi::xml_node root = xmlDoc.document_element();
-
-        auto version = GetIRVersion(root);
-        IRParser parser(version, extensions);
-        return parser.parse(root, weights);
-    } catch (const std::string& err) {
-        THROW_IE_EXCEPTION << err;
-    } catch (const details::InferenceEngineException& e) {
-        throw;
-    } catch (const std::exception& e) {
-        THROW_IE_EXCEPTION << e.what();
-    } catch (...) {
-        THROW_IE_EXCEPTION << "Unknown exception thrown";
-    }
-}
diff --git a/inference-engine/src/ir_readers/ie_ir_reader.hpp b/inference-engine/src/ir_readers/ie_ir_reader.hpp
deleted file mode 100644
index a2eb69880f6..00000000000
--- a/inference-engine/src/ir_readers/ie_ir_reader.hpp
+++ /dev/null
@@ -1,59 +0,0 @@
-// Copyright (C) 2018-2020 Intel Corporation
-// SPDX-License-Identifier: Apache-2.0
-//
-
-#pragma once
-
-#include <ie_api.h>
-#include <ie_blob.h>
-#include <ie_common.h>
-#include <ie_iextension.h>
-
-#include <ie_icnn_network.hpp>
-#include <map>
-#include <memory>
-#include <string>
-#include <vector>
-
-namespace pugi {
-class xml_node;
-class xml_document;
-}  // namespace pugi
-
-namespace ngraph {
-class Function;
-}  // namespace ngraph
-
-namespace InferenceEngine {
-
-/**
- * @brief This class is the main interface to build and parse a network from a given IR
- *
- * All methods here do not throw exceptions and return a StatusCode and ResponseDesc object.
- * Alternatively, to use methods that throw exceptions, refer to the CNNNetReader wrapper class.
- */
-class INFERENCE_ENGINE_API_CLASS(IRReader) {
-public:
-    IRReader() = default;
-    explicit IRReader(const std::vector<IExtensionPtr>& exts): extensions(exts) {}
-    /**
-     * @brief Reads IR xml and bin files
-     * @param modelPath path to IR file
-     * @param binPath path to bin file
-     * @return shared pointer to nGraph function
-     */
-    std::shared_ptr<ngraph::Function> read(const std::string& modelPath, const std::string& binPath = "");
-    /**
-     * @brief Reads IR xml and bin (with the same name) files
-     * @param model string with IR
-     * @param weights shared pointer to constant blob with weights
-     * @return shared pointer to nGraph function
-     */
-    std::shared_ptr<ngraph::Function> read(const std::string& model, const Blob::CPtr& weights);
-
-private:
-    std::shared_ptr<ngraph::Function> readXml(const pugi::xml_document& xmlDoc, const Blob::CPtr& weights);
-    std::vector<IExtensionPtr> extensions;
-};
-
-}  // namespace InferenceEngine
diff --git a/inference-engine/src/legacy_api/include/graph_tools.hpp b/inference-engine/src/legacy_api/include/graph_tools.hpp
index 614e1876165..3709dbb2a61 100644
--- a/inference-engine/src/legacy_api/include/graph_tools.hpp
+++ b/inference-engine/src/legacy_api/include/graph_tools.hpp
@@ -189,9 +189,10 @@ inline void UnorderedDFS(std::unordered_set<CNNLayer*>& visited, const Inference
         }
 
         // visit parents
-        for (auto&& input : cnnLayer->insData) {
+        for (size_t i = 0; i < cnnLayer->insData.size(); i++) {
+            auto& input = cnnLayer->insData[i];
             if (!input.lock()) {
-                THROW_IE_EXCEPTION << "Data inserted into layer " << cnnLayer->name << " is nullptr";
+                THROW_IE_EXCEPTION << "Data " << i << " inserted into layer " << cnnLayer->name << " is nullptr";
             } else {
                 auto creatorLayer = input.lock()->getCreatorLayer().lock();
                 if (creatorLayer) {
diff --git a/inference-engine/src/legacy_api/src/ie_cnn_layer_builder_ngraph.h b/inference-engine/src/legacy_api/src/ie_cnn_layer_builder_ngraph.h
index 4b45eda00e8..dae481a5356 100644
--- a/inference-engine/src/legacy_api/src/ie_cnn_layer_builder_ngraph.h
+++ b/inference-engine/src/legacy_api/src/ie_cnn_layer_builder_ngraph.h
@@ -61,7 +61,7 @@ public:
     CNNLayer::Ptr createLayer(const std::shared_ptr<ngraph::Node>& layer) const override;
 
     bool canCreate(const std::shared_ptr<ngraph::Node>& node) const override {
-        auto castedPtr = std::dynamic_pointer_cast<NGT>(node);
+        auto castedPtr = ngraph::as_type_ptr<NGT>(node);
         return castedPtr != nullptr;
     }
 
diff --git a/inference-engine/src/plugin_api/generic_ie.hpp b/inference-engine/src/plugin_api/generic_ie.hpp
index fbddd6856a1..7cb707aac6a 100644
--- a/inference-engine/src/plugin_api/generic_ie.hpp
+++ b/inference-engine/src/plugin_api/generic_ie.hpp
@@ -37,6 +37,8 @@ public:
             }
         }
         explicit DisableReshape(const std::shared_ptr<const ngraph::Function>& graph) {
+            IE_ASSERT(graph);
+
             for (auto& op : graph->get_ops()) {
                 addOp(op);
             }
diff --git a/inference-engine/src/readers/CMakeLists.txt b/inference-engine/src/readers/CMakeLists.txt
new file mode 100644
index 00000000000..fc03a338c93
--- /dev/null
+++ b/inference-engine/src/readers/CMakeLists.txt
@@ -0,0 +1,21 @@
+# Copyright (C) 2020 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+set(TARGET_NAME inference_engine_reader_api)
+
+# Reader API library
+add_library(${TARGET_NAME} INTERFACE)
+
+target_include_directories(${TARGET_NAME} INTERFACE
+    "${CMAKE_CURRENT_SOURCE_DIR}/reader_api"
+    $<TARGET_PROPERTY:inference_engine,INTERFACE_INCLUDE_DIRECTORIES>)
+
+file(GLOB_RECURSE reader_api_src "${CMAKE_CURRENT_SOURCE_DIR}/reader_api/*.hpp"
+                                 "${CMAKE_CURRENT_SOURCE_DIR}/reader_api/*.h")
+
+add_cpplint_target(${TARGET_NAME}_cpplint FOR_SOURCES ${reader_api_src})
+
+add_subdirectory(ir_reader)
+if(NGRAPH_ONNX_IMPORT_ENABLE)
+    add_subdirectory(onnx_reader)
+endif()
diff --git a/inference-engine/src/ir_readers/CMakeLists.txt b/inference-engine/src/readers/ir_reader/CMakeLists.txt
similarity index 89%
rename from inference-engine/src/ir_readers/CMakeLists.txt
rename to inference-engine/src/readers/ir_reader/CMakeLists.txt
index f90b2aeb185..6fc2110327f 100644
--- a/inference-engine/src/ir_readers/CMakeLists.txt
+++ b/inference-engine/src/readers/ir_reader/CMakeLists.txt
@@ -2,7 +2,7 @@
 # SPDX-License-Identifier: Apache-2.0
 #
 
-set(TARGET_NAME "inference_engine_ir_readers")
+set(TARGET_NAME "inference_engine_ir_reader")
 
 if(ENABLE_LTO)
     ie_enable_lto()
@@ -29,7 +29,7 @@ target_compile_definitions(${TARGET_NAME} PRIVATE IMPLEMENT_INFERENCE_ENGINE_API
 target_include_directories(${TARGET_NAME} PUBLIC ${PUBLIC_HEADERS_DIR})
 target_include_directories(${TARGET_NAME} PRIVATE "${IE_MAIN_SOURCE_DIR}/src/inference_engine")
 
-target_link_libraries(${TARGET_NAME} PUBLIC inference_engine_plugin_api ${NGRAPH_LIBRARIES} inference_engine)
+target_link_libraries(${TARGET_NAME} PUBLIC inference_engine_reader_api inference_engine_plugin_api ${NGRAPH_LIBRARIES} inference_engine)
 target_link_libraries(${TARGET_NAME} PRIVATE pugixml)
 
 # code style
diff --git a/inference-engine/src/ir_readers/ie_blob_proxy.hpp b/inference-engine/src/readers/ir_reader/ie_blob_proxy.hpp
similarity index 100%
rename from inference-engine/src/ir_readers/ie_blob_proxy.hpp
rename to inference-engine/src/readers/ir_reader/ie_blob_proxy.hpp
diff --git a/inference-engine/src/ir_readers/ie_cnn_net_reader_impl.cpp b/inference-engine/src/readers/ir_reader/ie_cnn_net_reader_impl.cpp
similarity index 85%
rename from inference-engine/src/ir_readers/ie_cnn_net_reader_impl.cpp
rename to inference-engine/src/readers/ir_reader/ie_cnn_net_reader_impl.cpp
index 7471390153a..b8364255126 100644
--- a/inference-engine/src/ir_readers/ie_cnn_net_reader_impl.cpp
+++ b/inference-engine/src/readers/ir_reader/ie_cnn_net_reader_impl.cpp
@@ -4,6 +4,7 @@
 
 #include <file_utils.h>
 #include <ie_cnn_net_reader_impl.h>
+#include <ie_blob_stream.hpp>
 
 #include <fstream>
 #include <map>
@@ -16,7 +17,7 @@
 #include "cnn_network_ngraph_impl.hpp"
 #include "details/os/os_filesystem.hpp"
 #include "ie_format_parser.h"
-#include "ie_ir_reader.hpp"
+#include "ie_ir_parser.hpp"
 #include "ie_profiling.hpp"
 #include "ie_plugin.hpp"
 #include "parsers.h"
@@ -39,10 +40,10 @@ StatusCode CNNNetReaderImpl::SetWeights(const TBlob<uint8_t>::Ptr& weights, Resp
     try {
         if (_version == 10) {
             // It's time to perform actual reading of V10 network and instantiate CNNNetworkNGraphImpl
-            IRReader v10Reader(extensions);
-            std::stringstream model;
-            xmlDoc->save(model);
-            network = std::make_shared<CNNNetworkNGraphImpl>(v10Reader.read(model.str(), weights));
+            IRParser parser(_version, extensions);
+            pugi::xml_node root = xmlDoc->document_element();
+            details::BlobStream blobStream(weights);
+            network = parser.parse(root, blobStream);
         } else if (weights) {
             _parser->SetWeights(weights);
         }
@@ -122,13 +123,9 @@ StatusCode CNNNetReaderImpl::ReadNetwork(const char* filepath, ResponseDesc* res
     return OK;
 }
 
-StatusCode CNNNetReaderImpl::ReadNetwork() {
-    description.clear();
-
+StatusCode CNNNetReaderImpl::ReadNetwork(const pugi::xml_node& const_root, ResponseDesc * desc) {
     try {
-        // check which version it is...
-        pugi::xml_node root = xmlDoc->document_element();
-
+        pugi::xml_node root = *const_cast<pugi::xml_node*>(&const_root);
         _version = GetFileVersion(root);
         if (_version < 2) THROW_IE_EXCEPTION << "deprecated IR version: " << _version;
         if (_version == 10) {
@@ -147,13 +144,36 @@ StatusCode CNNNetReaderImpl::ReadNetwork() {
             THROW_IE_EXCEPTION << "cannot parse future versions: " << _version;
         }
     } catch (const std::string& err) {
-        description = err;
         parseSuccess = false;
-        return GENERAL_ERROR;
+        return DescriptionBuffer(desc) << err;
     } catch (const InferenceEngineException& e) {
         description = e.what();
         parseSuccess = false;
-        return GENERAL_ERROR;
+        return DescriptionBuffer(desc) << e.what();
+    } catch (const std::exception& e) {
+        description = e.what();
+        parseSuccess = false;
+        return DescriptionBuffer(desc) << e.what();
+    } catch (...) {
+        parseSuccess = false;
+        return DescriptionBuffer(UNEXPECTED, desc) << "Unknown exception thrown";
+    }
+
+    return OK;
+}
+
+StatusCode CNNNetReaderImpl::ReadNetwork() {
+    description.clear();
+
+    try {
+        // check which version it is...
+        pugi::xml_node root = xmlDoc->document_element();
+
+        ResponseDesc resp;
+        StatusCode ret = ReadNetwork(root, &resp);
+        if (ret != OK)
+            description = resp.msg;
+        return ret;
     } catch (const std::exception& e) {
         description = e.what();
         parseSuccess = false;
diff --git a/inference-engine/src/ir_readers/ie_cnn_net_reader_impl.h b/inference-engine/src/readers/ir_reader/ie_cnn_net_reader_impl.h
similarity index 95%
rename from inference-engine/src/ir_readers/ie_cnn_net_reader_impl.h
rename to inference-engine/src/readers/ir_reader/ie_cnn_net_reader_impl.h
index bd0e0c59493..a291055e0d5 100644
--- a/inference-engine/src/ir_readers/ie_cnn_net_reader_impl.h
+++ b/inference-engine/src/readers/ir_reader/ie_cnn_net_reader_impl.h
@@ -44,6 +44,8 @@ public:
 
     StatusCode ReadNetwork(const void* model, size_t size, ResponseDesc* resp) noexcept override;
 
+    StatusCode ReadNetwork(const pugi::xml_node& root, ResponseDesc* resp);
+
     StatusCode SetWeights(const TBlob<uint8_t>::Ptr& weights, ResponseDesc* resp) noexcept override;
 
     StatusCode ReadWeights(const char* filepath, ResponseDesc* resp) noexcept override;
@@ -53,6 +55,10 @@ public:
         return network.get();
     }
 
+    std::shared_ptr<ICNNNetwork> getNetwork() {
+        return network;
+    }
+
     bool isParseSuccess(ResponseDesc* resp) noexcept override {
         return parseSuccess;
     }
diff --git a/inference-engine/src/ir_readers/ie_format_parser.cpp b/inference-engine/src/readers/ir_reader/ie_format_parser.cpp
similarity index 100%
rename from inference-engine/src/ir_readers/ie_format_parser.cpp
rename to inference-engine/src/readers/ir_reader/ie_format_parser.cpp
diff --git a/inference-engine/src/ir_readers/ie_format_parser.h b/inference-engine/src/readers/ir_reader/ie_format_parser.h
similarity index 100%
rename from inference-engine/src/ir_readers/ie_format_parser.h
rename to inference-engine/src/readers/ir_reader/ie_format_parser.h
diff --git a/inference-engine/src/ir_readers/ie_ir_parser.cpp b/inference-engine/src/readers/ir_reader/ie_ir_parser.cpp
similarity index 93%
rename from inference-engine/src/ir_readers/ie_ir_parser.cpp
rename to inference-engine/src/readers/ir_reader/ie_ir_parser.cpp
index 4eec2cce76a..4315d16f652 100644
--- a/inference-engine/src/ir_readers/ie_ir_parser.cpp
+++ b/inference-engine/src/readers/ir_reader/ie_ir_parser.cpp
@@ -6,6 +6,7 @@
 
 #include <ie_memcpy.h>
 
+#include <typeinfo>
 #include <unordered_set>
 #include <algorithm>
 #include <deque>
@@ -24,6 +25,8 @@
 #include <ngraph/opsets/opset3.hpp>
 #include <ngraph/variant.hpp>
 
+#include <cpp/ie_cnn_network.h>
+#include "ie_blob_stream.hpp"
 #include "cnn_network_impl.hpp"
 #include "details/caseless.hpp"
 #include "details/ie_cnn_network_tools.h"
@@ -32,12 +35,17 @@
 #include "generic_ie.hpp"
 #include "precision_utils.h"
 #include "blob_factory.hpp"
+#include "ie_cnn_net_reader_impl.h"
 
 using namespace InferenceEngine;
 using namespace XMLParseUtils;
 
 IRParser::IRParser(size_t version): IRParser(version, {}) {}
 IRParser::IRParser(size_t version, const std::vector<InferenceEngine::IExtensionPtr>& exts) {
+    if (version < 10) {
+        parser = std::make_shared<CNNParser>();
+        return;
+    }
     switch (version) {
     case 10:
         parser = std::make_shared<V10Parser>(exts);
@@ -47,8 +55,60 @@ IRParser::IRParser(size_t version, const std::vector<InferenceEngine::IExtension
     }
 }
 
-std::shared_ptr<ngraph::Function> IRParser::parse(const pugi::xml_node& root, const Blob::CPtr& weights) {
-    return parser->parse(root, weights);
+std::shared_ptr<ICNNNetwork> IRParser::parse(const pugi::xml_node& root, std::istream& binStream) {
+    return parser->parse(root, binStream);
+}
+
+/**
+ * Hold original blob in order to avoid situations when original blob is allocated on stack
+ */
+class WeightsHolderBlob : public TBlob<uint8_t> {
+    Blob::CPtr originBlob;
+
+public:
+    explicit WeightsHolderBlob(const Blob::CPtr& weights) :
+        TBlob<uint8_t>(weights->getTensorDesc(),
+                       weights->cbuffer().as<uint8_t*>()),
+        originBlob(weights) { }
+};
+
+std::shared_ptr<ICNNNetwork> CNNParser::parse(const pugi::xml_node& root, std::istream& binStream) {
+    auto getBlobStream = [](std::istream& binStream) {
+        details::BlobStream* blobStream = dynamic_cast<details::BlobStream*>(&binStream);
+        if (blobStream == nullptr) {
+            details::BlobStream helper({});
+            std::string typeStream = typeid(binStream).name();
+            std::string typeBlobStream = typeid(helper).name();
+            if (typeStream == typeBlobStream)
+                blobStream = static_cast<details::BlobStream*>(&binStream);
+        }
+        return blobStream;
+    };
+    details::CNNNetReaderImpl reader(std::make_shared<details::V2FormatParserCreator>());
+    ResponseDesc resp;
+    StatusCode ret = reader.ReadNetwork(root, &resp);
+    if (ret != OK)
+        THROW_IE_EXCEPTION << resp.msg;
+    TBlob<uint8_t>::Ptr weightsPtr;
+
+    // Try to get BlobStream to work with original blob
+    details::BlobStream* blobStream = getBlobStream(binStream);
+    if (blobStream != nullptr) {
+        weightsPtr = std::make_shared<WeightsHolderBlob>(blobStream->getBlob());
+    } else {
+        // Allocate a blob for weights
+        binStream.seekg(0, std::ios::end);
+        size_t length = binStream.tellg();
+        weightsPtr = std::make_shared<TBlob<uint8_t>>(TensorDesc(Precision::U8, {length}, Layout::C));
+        weightsPtr->allocate();
+        char* data = weightsPtr->buffer().as<char*>();
+        binStream.seekg(0, std::ios::beg);
+        binStream.read(data, length);
+    }
+    ret = reader.SetWeights(weightsPtr, &resp);
+    if (ret != OK)
+        THROW_IE_EXCEPTION << resp.msg;
+    return reader.getNetwork();
 }
 
 V10Parser::V10Parser(const std::vector<IExtensionPtr>& exts) {
@@ -68,7 +128,7 @@ V10Parser::V10Parser(const std::vector<IExtensionPtr>& exts) {
     }
 }
 
-std::shared_ptr<ngraph::Function> V10Parser::parse(const pugi::xml_node& root, const Blob::CPtr& weights) {
+std::shared_ptr<ICNNNetwork> V10Parser::parse(const pugi::xml_node& root, std::istream& binStream) {
     using node_params = struct {
         pugi::xml_node xml;
         GenericLayerParams params;
@@ -139,7 +199,7 @@ std::shared_ptr<ngraph::Function> V10Parser::parse(const pugi::xml_node& root, c
                 input_node->output(p_output.getRealOutputPortId(e.fromPortId));
         }
 
-        auto node = createNode(inputs, p.xml, weights, p.params);
+        auto node = createNode(inputs, p.xml, binStream, p.params);
         id_to_node[layer_id] = node;
 
         // Check that output shape after nGraph node validation the same as in IR
@@ -182,7 +242,7 @@ std::shared_ptr<ngraph::Function> V10Parser::parse(const pugi::xml_node& root, c
             result_nodes[0]->add_control_dependency(assign);
         }
     }
-    return function;
+    return CNNNetwork(function);
 }
 
 V10Parser::GenericLayerParams V10Parser::parseGenericParams(const pugi::xml_node& node) {
@@ -243,7 +303,7 @@ bool V10Parser::LayerBaseCreator::shouldCreate(const std::string& nodeType) cons
 }
 
 std::shared_ptr<ngraph::Node> V10Parser::createNode(const std::vector<ngraph::Output<ngraph::Node>>& inputs,
-                                                    const pugi::xml_node& node, const Blob::CPtr& weights,
+                                                    const pugi::xml_node& node, std::istream& binStream,
                                                     const GenericLayerParams& params) {
     static std::vector<std::shared_ptr<LayerBaseCreator>> creators = {
         std::make_shared<LayerCreator<ngraph::op::Abs>>("Abs"),
@@ -376,7 +436,7 @@ std::shared_ptr<ngraph::Node> V10Parser::createNode(const std::vector<ngraph::Ou
                 useCreator |= opset.contains_type(creator->getNodeType()) || !opset.contains_type(params.type);
             }
             if (useCreator)
-                ngraphNode = creator->createLayer(inputs, node, weights, params);
+                ngraphNode = creator->createLayer(inputs, node, binStream, params);
             break;
         }
     }
@@ -409,6 +469,9 @@ std::shared_ptr<ngraph::Node> V10Parser::createNode(const std::vector<ngraph::Ou
 
         auto blobs = node.child("blobs");
         if (!blobs.empty()) {
+            binStream.seekg(0, std::ios::end);
+            std::streampos length = binStream.tellg();
+
             for (pugi::xml_node blob = blobs.first_child(); !blob.empty(); blob = blob.next_sibling()) {
                 size_t size = GetUInt64Attr(blob, "size", 0);
                 uint64_t offset = GetUInt64Attr(blob, "offset", 0);
@@ -417,16 +480,18 @@ std::shared_ptr<ngraph::Node> V10Parser::createNode(const std::vector<ngraph::Ou
                 if (!preStr.empty())
                     precision = Precision::FromStr(preStr);
                 if (!size) continue;
-                if (!weights || weights->cbuffer() == nullptr)
+                if (!length)
                     THROW_IE_EXCEPTION << "Cannot read network! The model requires weights data! "
                         << "Bin file cannot be found! Please specify the path to bin file.";
-                if (weights->byteSize() < offset + size)
+                if (length < offset + size)
                     THROW_IE_EXCEPTION << "Cannot create " << params.type << " layer with name: " << params.name
                                        << ". Layer has incorrect weights!";
-                const uint8_t* data = weights->cbuffer().as<uint8_t*>() + offset;
                 Blob::Ptr wBlob = make_blob_with_precision(TensorDesc(precision, {size / precision.size()}, Layout::C));
                 wBlob->allocate();
-                ie_memcpy(wBlob->buffer().as<uint8_t*>(), size, data, size);
+                char* data = wBlob->buffer().as<char*>();
+                binStream.seekg(offset, std::ios::beg);
+                binStream.read(data, size);
+
                 Blob::CPtr cBlob = wBlob;
                 parameters[blob.name()] = wBlob;
             }
@@ -467,7 +532,7 @@ namespace InferenceEngine {
 // DetectionOutput layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::DetectionOutput>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
@@ -515,7 +580,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::DetectionOutpu
 // TensorIterator layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::TensorIterator>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     auto tensor_iterator = std::make_shared<ngraph::op::TensorIterator>();
     tensor_iterator->set_friendly_name(GetStrAttr(node, "name"));
@@ -539,7 +604,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::TensorIterator
 
     // Create ngraph::Function, convert it to ngraph::BodyLambda and set it as TensorIterator body
     IRParser parser(10);
-    auto ngraph_function = parser.parse(node.child("body"), weights);
+    auto ngraph_function = parser.parse(node.child("body"), binStream)->getFunction();
     auto parameter_nodes = ngraph_function->get_parameters();
     auto result_nodes = ngraph_function->get_results();
     // Disabled reshape for generic operations in the TI body
@@ -660,7 +725,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::TensorIterator
 // PriorBoxClustered layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PriorBoxClustered>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -688,7 +753,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PriorBoxCluste
 // Proposal layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Proposal>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     pugi::xml_node dn = node.child("data");
@@ -718,7 +783,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Proposal>::cre
 // PriorBox layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PriorBox>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -746,7 +811,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PriorBox>::cre
 // ShapeOf layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ShapeOf>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::ShapeOf>(inputs[0]);
@@ -755,7 +820,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ShapeOf>::crea
 // FakeQuantize layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::FakeQuantize>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 5);
     pugi::xml_node dn = node.child("data");
@@ -769,7 +834,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::FakeQuantize>:
 // ReverseSequence layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ReverseSequence>::createLayer(const ngraph::OutputVector & inputs, const pugi::xml_node& node,
-                                                                                                const Blob::CPtr& weights,
+                                                                                                std::istream& binStream,
                                                                                                 const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -779,7 +844,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ReverseSequenc
 // Covnert layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Convert>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -793,7 +858,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Convert>::crea
 // LSTMCell layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::LSTMCell>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 6);
     pugi::xml_node dn = node.child("data");
@@ -812,7 +877,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::LSTMCell>::cre
 // BatchNormInference layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::BatchNormInference>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 5);
     pugi::xml_node dn = node.child("data");
@@ -826,7 +891,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::BatchNormInfer
 // CTCGreedyDecoder layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::CTCGreedyDecoder>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -840,7 +905,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::CTCGreedyDecod
 // TopK layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::TopK>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -877,7 +942,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::TopK>::cre
 // Pad layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Pad>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
@@ -911,7 +976,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Pad>::crea
 // SquaredDifference layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::SquaredDifference>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::SquaredDifference>(inputs[0], inputs[1]);
@@ -920,7 +985,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::SquaredDiffere
 // Greater layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Greater>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Greater>(inputs[0], inputs[1]);
@@ -929,7 +994,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Greater>::
 // GreaterEqual layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GreaterEqual>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::GreaterEqual>(inputs[0], inputs[1]);
@@ -938,7 +1003,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GreaterEqu
 // Less layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Less>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Less>(inputs[0], inputs[1]);
@@ -947,7 +1012,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Less>::cre
 // LessEqual layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LessEqual>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::LessEqual>(inputs[0], inputs[1]);
@@ -956,7 +1021,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LessEqual>
 // Equal layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Equal>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Equal>(inputs[0], inputs[1]);
@@ -965,7 +1030,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Equal>::cr
 // NotEqual layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::NotEqual>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::NotEqual>(inputs[0], inputs[1]);
@@ -974,7 +1039,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::NotEqual>:
 // FloorMod layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::FloorMod>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::FloorMod>(inputs[0], inputs[1]);
@@ -983,7 +1048,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::FloorMod>:
 // Select layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Select>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     return std::make_shared<ngraph::op::v1::Select>(inputs[0], inputs[1], inputs[2]);
@@ -992,7 +1057,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Select>::c
 // MVN layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::MVN>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1009,7 +1074,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::MVN>::createLa
 // Log layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Log>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Log>(inputs[0]);
@@ -1018,7 +1083,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Log>::createLa
 // LRN layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::LRN>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1036,7 +1101,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::LRN>::createLa
 // Clamp layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Clamp>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1052,7 +1117,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Clamp>::create
 // VariadicSplit layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::VariadicSplit>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     return std::make_shared<ngraph::op::VariadicSplit>(inputs[0], inputs[1], inputs[2]);
@@ -1061,7 +1126,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::VariadicSplit>
 // Split layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Split>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
@@ -1076,7 +1141,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Split>::cr
 // Sigmoid layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sigmoid>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Sigmoid>(inputs[0]);
@@ -1085,7 +1150,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sigmoid>::crea
 // ELU layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Elu>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1099,7 +1164,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Elu>::createLa
 // SpaceToDepth layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::SpaceToDepth>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1113,7 +1178,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::SpaceToDepth>:
 // DepthToSpace layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::DepthToSpace>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1127,7 +1192,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::DepthToSpace>:
 // SeLU layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v0::Selu>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     return std::make_shared<ngraph::op::v0::Selu>(inputs[0], inputs[1], inputs[2]);
@@ -1136,7 +1201,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v0::Selu>::cre
 // PReLU layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PRelu>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::PRelu>(inputs[0], inputs[1]);
@@ -1145,7 +1210,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PRelu>::create
 // Exp layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Exp>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Exp>(inputs[0]);
@@ -1154,7 +1219,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Exp>::createLa
 // ReLU layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Relu>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Relu>(inputs[0]);
@@ -1163,7 +1228,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Relu>::createL
 // Negative layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Negative>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Negative>(inputs[0]);
@@ -1172,7 +1237,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Negative>::cre
 // Range layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Range>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     return std::make_shared<ngraph::op::Range>(inputs[0], inputs[1], inputs[2]);
@@ -1181,7 +1246,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Range>::create
 // Tanh layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Tanh>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Tanh>(inputs[0]);
@@ -1190,7 +1255,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Tanh>::createL
 // Result layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Result>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Result>(inputs[0]);
@@ -1199,7 +1264,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Result>::creat
 // Tile layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Tile>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::Tile>(inputs[0], inputs[1]);
@@ -1208,7 +1273,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Tile>::createL
 // StridedSlice layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::StridedSlice>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
 
     pugi::xml_node dn = node.child("data");
@@ -1233,7 +1298,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::StridedSli
 // Reshape layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Reshape>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
 
@@ -1247,7 +1312,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Reshape>::
 // Squeeze layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Squeeze>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::Squeeze>(inputs[0], inputs[1]);
@@ -1256,7 +1321,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Squeeze>::crea
 // Unsqueeze layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Unsqueeze>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::Unsqueeze>(inputs[0], inputs[1]);
@@ -1265,7 +1330,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Unsqueeze>::cr
 // Interpolate layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Interpolate>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
 
@@ -1299,7 +1364,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Interpolate>::
 // Abs layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Abs>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Abs>(inputs[0]);
@@ -1308,7 +1373,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Abs>::createLa
 // Add layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Add>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Add>(inputs[0], inputs[1]);
@@ -1317,7 +1382,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Add>::crea
 // Minimum layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Minimum>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Minimum>(inputs[0], inputs[1]);
@@ -1326,7 +1391,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Minimum>::
 // Maximum layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Maximum>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Maximum>(inputs[0], inputs[1]);
@@ -1335,7 +1400,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Maximum>::
 // Divide layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Divide>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Divide>(inputs[0], inputs[1]);
@@ -1344,7 +1409,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Divide>::c
 // Subtract layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Subtract>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Subtract>(inputs[0], inputs[1]);
@@ -1353,7 +1418,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Subtract>:
 // Multiply layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Multiply>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Multiply>(inputs[0], inputs[1]);
@@ -1362,7 +1427,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Multiply>:
 // Broadcast layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Broadcast>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     if (inputs.size() == 2) {
         return std::make_shared<ngraph::op::v1::Broadcast>(inputs[0], inputs[1]);
@@ -1375,7 +1440,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Broadcast>
 // Constant layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Constant>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 0);
     pugi::xml_node dn = node.child("data");
@@ -1386,10 +1451,12 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Constant>::cre
     size_t offset = GetUInt64Attr(dn, "offset");
     size_t size = GetUInt64Attr(dn, "size");
 
-    if (!weights || weights->cbuffer() == nullptr)
+    binStream.seekg(0, std::ios::end);
+    std::streampos length = binStream.tellg();
+    if (!length)
         THROW_IE_EXCEPTION << "Cannot read network! The model requires weights data! "
             << "Bin file cannot be found! Please specify the path to bin file.";
-    if (weights->byteSize() < offset + size)
+    if (length < offset + size)
         THROW_IE_EXCEPTION << "Cannot create " << getType() << " layer with name: " << layerParsePrms.name
                            << ". Layer has incorrect weights!";
 
@@ -1399,15 +1466,17 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Constant>::cre
     if (size < std::ceil(ngraph::shape_size(shape) * el_type.bitwidth() / 8.f))
         THROW_IE_EXCEPTION << "Cannot create Constant op " << layerParsePrms.name << " size attribute and shape size are inconsistent!";
 
-    const void * data = (weights->cbuffer().as<uint8_t *>() + offset);
-
-    return std::make_shared<ngraph::op::Constant>(port.precision, shape, data);
+    auto constant = std::make_shared<ngraph::op::Constant>(port.precision, shape);
+    char* data = const_cast<char*>(reinterpret_cast<const char*>(constant->get_data_ptr()));
+    binStream.seekg(offset, std::ios::beg);
+    binStream.read(data, size);
+    return constant;
 }
 
 // Power layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Power>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::Power>(inputs[0], inputs[1]);
@@ -1416,7 +1485,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Power>::cr
 // MatMul layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::MatMul>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1430,7 +1499,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::MatMul>::creat
 // Softmax layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Softmax>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1444,7 +1513,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Softmax>::
 // Sqrt layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sqrt>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Sqrt>(inputs[0]);
@@ -1453,7 +1522,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sqrt>::createL
 // RegionYolo layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::RegionYolo>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1477,7 +1546,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::RegionYolo>::c
 // ReorgYolo layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ReorgYolo>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1492,7 +1561,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ReorgYolo>::cr
 // ReduceMin layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceMin>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1506,7 +1575,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceMin>
 // ReduceMax layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceMax>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1520,7 +1589,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceMax>
 // ReduceMean layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceMean>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1534,7 +1603,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceMean
 // ReduceProd layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceProd>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1548,7 +1617,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceProd
 // ReduceSum layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceSum>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1562,7 +1631,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceSum>
 // Transpose layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Transpose>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::Transpose>(inputs[0], inputs[1]);
@@ -1571,7 +1640,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Transpose>::cr
 // BinaryConvolution layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::BinaryConvolution>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1606,7 +1675,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::BinaryConv
 // Convolution layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Convolution>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1636,7 +1705,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Convolutio
 // GroupConvolution layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GroupConvolution>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1666,7 +1735,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GroupConvo
 // DeformableConvolution layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::DeformableConvolution>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     pugi::xml_node dn = node.child("data");
@@ -1699,7 +1768,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Deformable
 // ConvolutionBackpropData layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ConvolutionBackpropData>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
@@ -1737,7 +1806,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Convolutio
 // GroupConvolutionBackpropData layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GroupConvolutionBackpropData>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
@@ -1776,7 +1845,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GroupConvo
 // AvgPool layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::AvgPool>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1817,7 +1886,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::AvgPool>::
 // MaxPool layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::MaxPool>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -1857,7 +1926,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::MaxPool>::
 // ROIPooling layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ROIPooling>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1876,7 +1945,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::ROIPooling>::c
 // PSROIPooling layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PSROIPooling>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -1900,7 +1969,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::PSROIPooling>:
 
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::DeformablePSROIPooling>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
@@ -1935,7 +2004,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Deformable
 // Concat layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Concat>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, -1);
     pugi::xml_node dn = node.child("data");
@@ -1949,7 +2018,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Concat>::creat
 // Gather layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Gather>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     return std::make_shared<ngraph::op::v1::Gather>(inputs[0], inputs[1], inputs[2]);
@@ -1958,7 +2027,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::Gather>::c
 // GatherTree layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GatherTree>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 4);
     return std::make_shared<ngraph::op::v1::GatherTree>(inputs[0], inputs[1], inputs[2], inputs[3]);
@@ -1967,7 +2036,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::GatherTree
 // OneHot layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::OneHot>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 4);
 
@@ -1981,7 +2050,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::OneHot>::c
 // NormalizeL2 layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::NormalizeL2>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -2006,7 +2075,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::NormalizeL2>::
 // Erf layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Erf>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Erf>(inputs[0]);
@@ -2015,7 +2084,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Erf>::createLa
 // Sin layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sin>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Sin>(inputs[0]);
@@ -2024,7 +2093,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sin>::createLa
 // Sign layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sign>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Sign>(inputs[0]);
@@ -2033,7 +2102,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sign>::createL
 // Sinh layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sinh>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Sinh>(inputs[0]);
@@ -2042,7 +2111,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Sinh>::createL
 // Asin layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Asin>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Asin>(inputs[0]);
@@ -2051,7 +2120,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Asin>::createL
 // Cos layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Cos>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Cos>(inputs[0]);
@@ -2060,7 +2129,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Cos>::createLa
 // Cosh layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Cosh>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Cosh>(inputs[0]);
@@ -2069,7 +2138,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Cosh>::createL
 // Acos layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Acos>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Acos>(inputs[0]);
@@ -2078,7 +2147,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Acos>::createL
 // Tan layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Tan>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Tan>(inputs[0]);
@@ -2087,7 +2156,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Tan>::createLa
 // Atan layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Atan>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Atan>(inputs[0]);
@@ -2096,7 +2165,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Atan>::createL
 // Floor layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Floor>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Floor>(inputs[0]);
@@ -2105,7 +2174,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Floor>::create
 // Ceiling layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Ceiling>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::Ceiling>(inputs[0]);
@@ -2114,7 +2183,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::Ceiling>::crea
 // HardSigmoid layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::HardSigmoid>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 3);
     return std::make_shared<ngraph::op::HardSigmoid>(inputs[0], inputs[1], inputs[2]);
@@ -2123,7 +2192,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::HardSigmoid>::
 // GRN layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::GRN>::createLayer(
-    const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     pugi::xml_node dn = node.child("data");
@@ -2137,7 +2206,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::GRN>::createLa
 // LogicalAnd layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalAnd>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::LogicalAnd>(inputs[0], inputs[1]);
@@ -2146,7 +2215,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalAnd
 // LogicalOr layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalOr>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::LogicalOr>(inputs[0], inputs[1]);
@@ -2155,7 +2224,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalOr>
 // LogicalXor layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalXor>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     return std::make_shared<ngraph::op::v1::LogicalXor>(inputs[0], inputs[1]);
@@ -2164,7 +2233,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalXor
 // LogicalNot layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalNot>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 1);
     return std::make_shared<ngraph::op::v1::LogicalNot>(inputs[0]);
@@ -2173,7 +2242,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::LogicalNot
 // ReduceLogicalAnd layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceLogicalAnd>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -2187,7 +2256,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceLogi
 // ReduceLogicalOr layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceLogicalOr>::createLayer(
-    const ngraph::OutputVector & inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+    const ngraph::OutputVector & inputs, const pugi::xml_node& node, std::istream& binStream,
     const GenericLayerParams& layerParsePrms) {
     checkParameters(inputs, layerParsePrms, 2);
     pugi::xml_node dn = node.child("data");
@@ -2201,7 +2270,7 @@ std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::ReduceLogi
 // NonMaxSuppression layer
 template <>
 std::shared_ptr<ngraph::Node> V10Parser::LayerCreator<ngraph::op::v1::NonMaxSuppression>::createLayer(
-        const ngraph::OutputVector& inputs, const pugi::xml_node& node, const Blob::CPtr& weights,
+        const ngraph::OutputVector& inputs, const pugi::xml_node& node, std::istream& binStream,
         const GenericLayerParams& layerParsePrms) {
     pugi::xml_node dn = node.child("data");
 
diff --git a/inference-engine/src/ir_readers/ie_ir_parser.hpp b/inference-engine/src/readers/ir_reader/ie_ir_parser.hpp
similarity index 93%
rename from inference-engine/src/ir_readers/ie_ir_parser.hpp
rename to inference-engine/src/readers/ir_reader/ie_ir_parser.hpp
index 17b331cb3fb..c783c2b1f58 100644
--- a/inference-engine/src/ir_readers/ie_ir_parser.hpp
+++ b/inference-engine/src/readers/ir_reader/ie_ir_parser.hpp
@@ -27,24 +27,30 @@ class IParser {
 public:
     using Ptr = std::shared_ptr<IParser>;
     virtual ~IParser() = default;
-    virtual std::shared_ptr<ngraph::Function> parse(const pugi::xml_node& root, const Blob::CPtr& weights) = 0;
+    virtual std::shared_ptr<ICNNNetwork> parse(const pugi::xml_node& root, std::istream& binStream) = 0;
 };
 
 class IRParser {
 public:
     explicit IRParser(size_t version);
     IRParser(size_t version, const std::vector<InferenceEngine::IExtensionPtr>& exts);
-    std::shared_ptr<ngraph::Function> parse(const pugi::xml_node& root, const Blob::CPtr& weights);
+    std::shared_ptr<ICNNNetwork> parse(const pugi::xml_node& root, std::istream& binStream);
     virtual ~IRParser() = default;
 
 private:
     IParser::Ptr parser;
 };
 
+class CNNParser : public IParser {
+public:
+    CNNParser() = default;
+    std::shared_ptr<ICNNNetwork> parse(const pugi::xml_node& root, std::istream& binStream) override;
+};
+
 class V10Parser : public IParser {
 public:
     explicit V10Parser(const std::vector<IExtensionPtr>& exts);
-    std::shared_ptr<ngraph::Function> parse(const pugi::xml_node& root, const Blob::CPtr& weights) override;
+    std::shared_ptr<ICNNNetwork> parse(const pugi::xml_node& root, std::istream& binStream) override;
 
 private:
     std::map<std::string, ngraph::OpSet> opsets;
@@ -136,7 +142,7 @@ private:
     public:
         virtual ~LayerBaseCreator() {}
         virtual std::shared_ptr<ngraph::Node> createLayer(const ngraph::OutputVector& inputs,
-                                                          const pugi::xml_node& node, const Blob::CPtr& weights,
+                                                          const pugi::xml_node& node, std::istream& binStream,
                                                           const GenericLayerParams& layerParsePrms) = 0;
 
         bool shouldCreate(const std::string& nodeType) const;
@@ -148,7 +154,7 @@ private:
     public:
         explicit LayerCreator(const std::string& type): LayerBaseCreator(type) {}
         std::shared_ptr<ngraph::Node> createLayer(const ngraph::OutputVector& inputs, const pugi::xml_node& node,
-                                                  const Blob::CPtr& weights,
+                                                  std::istream& binStream,
                                                   const GenericLayerParams& layerParsePrms) override;
         ngraph::NodeTypeInfo getNodeType() const override {
             return T::type_info;
@@ -156,7 +162,7 @@ private:
     };
 
     std::shared_ptr<ngraph::Node> createNode(const ngraph::OutputVector& inputs, const pugi::xml_node& node,
-                                             const Blob::CPtr& weights, const GenericLayerParams& params);
+                                             std::istream& binStream, const GenericLayerParams& params);
 
     GenericLayerParams parseGenericParams(const pugi::xml_node& node);
 
diff --git a/inference-engine/src/readers/ir_reader/ie_ir_reader.cpp b/inference-engine/src/readers/ir_reader/ie_ir_reader.cpp
new file mode 100644
index 00000000000..0abb43dfb46
--- /dev/null
+++ b/inference-engine/src/readers/ir_reader/ie_ir_reader.cpp
@@ -0,0 +1,61 @@
+// Copyright (C) 2018-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include <file_utils.h>
+#include <xml_parse_utils.h>
+
+#include <ie_ir_reader.hpp>
+#include <memory>
+#include <ngraph/ngraph.hpp>
+#include <string>
+#include <vector>
+#include <sstream>
+
+#include "description_buffer.hpp"
+#include "ie_ir_parser.hpp"
+#include "ie_ngraph_utils.hpp"
+
+using namespace InferenceEngine;
+
+static size_t GetIRVersion(pugi::xml_node& root) {
+    return XMLParseUtils::GetUIntAttr(root, "version", 0);
+}
+
+bool IRReader::supportModel(std::istream& model) const {
+    model.seekg(0, model.beg);
+    const int header_size = 128;
+    std::string header(header_size, ' ');
+    model.read(&header[0], header_size);
+    // find '<net ' substring in the .xml file
+    return (header.find("<net ") != std::string::npos) || (header.find("<Net ") != std::string::npos);
+}
+
+CNNNetwork IRReader::read(std::istream& model, const std::vector<IExtensionPtr>& exts) const {
+    std::istringstream emptyStream;
+    return read(model, emptyStream, exts);
+}
+CNNNetwork IRReader::read(std::istream& model, std::istream& weights, const std::vector<IExtensionPtr>& exts) const {
+    model.seekg(0, model.beg);
+    weights.seekg(0, weights.beg);
+    pugi::xml_document xmlDoc;
+    pugi::xml_parse_result res = xmlDoc.load(model);
+    if (res.status != pugi::status_ok) {
+        THROW_IE_EXCEPTION << res.description() << "at offset " << res.offset;
+    }
+    pugi::xml_node root = xmlDoc.document_element();
+
+    auto version = GetIRVersion(root);
+    IRParser parser(version, exts);
+    return CNNNetwork(parser.parse(root, weights));
+}
+
+INFERENCE_PLUGIN_API(StatusCode) InferenceEngine::CreateReader(IReader*& reader, ResponseDesc *resp) noexcept {
+    try {
+        reader = new IRReader();
+        return OK;
+    }
+    catch (std::exception &ex) {
+        return GENERAL_ERROR;
+    }
+}
diff --git a/inference-engine/src/readers/ir_reader/ie_ir_reader.hpp b/inference-engine/src/readers/ir_reader/ie_ir_reader.hpp
new file mode 100644
index 00000000000..d8ddd040a27
--- /dev/null
+++ b/inference-engine/src/readers/ir_reader/ie_ir_reader.hpp
@@ -0,0 +1,72 @@
+// Copyright (C) 2018-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include <ie_api.h>
+#include <ie_blob.h>
+#include <ie_common.h>
+#include <ie_iextension.h>
+
+#include <ie_icnn_network.hpp>
+#include <ie_reader.hpp>
+#include <map>
+#include <memory>
+#include <string>
+#include <vector>
+
+namespace pugi {
+class xml_node;
+class xml_document;
+}  // namespace pugi
+
+namespace ngraph {
+class Function;
+}  // namespace ngraph
+
+namespace InferenceEngine {
+
+/**
+ * @brief This class is the main interface to build and parse a network from a given IR
+ *
+ * All methods here do not throw exceptions and return a StatusCode and ResponseDesc object.
+ * Alternatively, to use methods that throw exceptions, refer to the CNNNetReader wrapper class.
+ */
+class IRReader: public IReader {
+public:
+    IRReader() = default;
+
+    void Release() noexcept override {
+        delete this;
+    }
+    /**
+     * @brief Checks that reader supports format of the model
+     * @param model stream with model
+     * @return true if format is supported
+     */
+    bool supportModel(std::istream& model) const override;
+    /**
+     * @brief Reads the model to CNNNetwork
+     * @param model stream with model
+     * @param exts vector with extensions
+     *
+     * @return CNNNetwork
+     */
+    CNNNetwork read(std::istream& model, const std::vector<IExtensionPtr>& exts) const override;
+    /**
+     * @brief Reads the model to CNNNetwork
+     * @param model stream with model
+     * @param weights stream with binary data
+     * @param exts vector with extensions
+     *
+     * @return CNNNetwork
+     */
+    CNNNetwork read(std::istream& model, std::istream& weights, const std::vector<IExtensionPtr>& exts) const override;
+
+    std::vector<std::string> getDataFileExtensions() const override {
+        return {"bin"};
+    }
+};
+
+}  // namespace InferenceEngine
diff --git a/inference-engine/src/ir_readers/ie_layer_parsers.cpp b/inference-engine/src/readers/ir_reader/ie_layer_parsers.cpp
similarity index 94%
rename from inference-engine/src/ir_readers/ie_layer_parsers.cpp
rename to inference-engine/src/readers/ir_reader/ie_layer_parsers.cpp
index 1855bbb1232..990e58d8ae0 100644
--- a/inference-engine/src/ir_readers/ie_layer_parsers.cpp
+++ b/inference-engine/src/readers/ir_reader/ie_layer_parsers.cpp
@@ -138,6 +138,18 @@ public:
         net->getOutputsInfo(out_info_map);
 
         IE_ASSERT(in_info_map.size() == inputs.size()) << "TI body. There are unlinked inputs";
+        for (auto& it : net->allLayers()) {
+            auto layer = it.second;
+            if (layer->type == "Input" || !layer->insData.empty())
+                continue;
+            if (!holder)
+                holder = std::make_shared<Data>("const_holder", Precision::UNSPECIFIED);
+            holder->getInputTo()[it.first] = layer;
+        }
+    }
+
+    DataPtr getHolder() {
+        return holder;
     }
 
     void setWeights(const WBlob& weights) {
@@ -155,6 +167,7 @@ private:
     pugi::xml_node& body;
     FormatParser parser;
     Precision default_precision;
+    DataPtr holder;
 
     PortMap inputs;
     PortMap outputs;
@@ -253,6 +266,8 @@ CNNLayer::Ptr TILayerCreator::CreateLayer(pugi::xml_node& node, LayerParseParame
 
     auto res = std::make_shared<TensorIterator>(layerParsePrms.prms);
     res->body.inputs = inputs;
+    if (auto holder = parser->getHolder())
+        res->body.inputs.emplace_back(holder);
     res->body.outputs = outputs;
     res->input_port_map = in_ports_maping;
     res->output_port_map = out_ports_maping;
diff --git a/inference-engine/src/ir_readers/ie_layer_parsers.h b/inference-engine/src/readers/ir_reader/ie_layer_parsers.h
similarity index 100%
rename from inference-engine/src/ir_readers/ie_layer_parsers.h
rename to inference-engine/src/readers/ir_reader/ie_layer_parsers.h
diff --git a/inference-engine/src/ir_readers/parsers.h b/inference-engine/src/readers/ir_reader/parsers.h
similarity index 100%
rename from inference-engine/src/ir_readers/parsers.h
rename to inference-engine/src/readers/ir_reader/parsers.h
diff --git a/inference-engine/src/readers/onnx_reader/CMakeLists.txt b/inference-engine/src/readers/onnx_reader/CMakeLists.txt
new file mode 100644
index 00000000000..17085811022
--- /dev/null
+++ b/inference-engine/src/readers/onnx_reader/CMakeLists.txt
@@ -0,0 +1,41 @@
+# Copyright (C) 2020 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+#
+
+set(TARGET_NAME "inference_engine_onnx_reader")
+
+if(ENABLE_LTO)
+    ie_enable_lto()
+endif()
+
+set(PUBLIC_HEADERS_DIR "${CMAKE_CURRENT_SOURCE_DIR}/")
+
+file(GLOB_RECURSE LIBRARY_SRC ${CMAKE_CURRENT_SOURCE_DIR}/*.cpp)
+file(GLOB_RECURSE PUBLIC_HEADERS ${PUBLIC_HEADERS_DIR}/*.h ${PUBLIC_HEADERS_DIR}/*.hpp)
+
+# Create named folders for the sources within the .vcproj
+# Empty name lists them directly under the .vcproj
+
+source_group("src" FILES ${LIBRARY_SRC})
+source_group("include" FILES ${PUBLIC_HEADERS})
+
+# Create shared library
+
+add_library(${TARGET_NAME} SHARED ${LIBRARY_SRC} ${PUBLIC_HEADERS})
+
+target_include_directories(${TARGET_NAME} PUBLIC ${PUBLIC_HEADERS_DIR})
+
+target_compile_definitions(${TARGET_NAME} PRIVATE IMPLEMENT_INFERENCE_ENGINE_PLUGIN)
+target_link_libraries(${TARGET_NAME} PRIVATE inference_engine_reader_api onnx_importer inference_engine)
+
+# code style
+
+add_cpplint_target(${TARGET_NAME}_cpplint FOR_TARGETS ${TARGET_NAME})
+
+# install
+
+install(TARGETS ${TARGET_NAME}
+        RUNTIME DESTINATION ${IE_CPACK_LIBRARY_PATH} COMPONENT core
+        ARCHIVE DESTINATION ${IE_CPACK_LIBRARY_PATH} COMPONENT core
+        LIBRARY DESTINATION ${IE_CPACK_LIBRARY_PATH} COMPONENT core)
+
diff --git a/inference-engine/src/readers/onnx_reader/ie_onnx_reader.cpp b/inference-engine/src/readers/onnx_reader/ie_onnx_reader.cpp
new file mode 100644
index 00000000000..fee5cca5251
--- /dev/null
+++ b/inference-engine/src/readers/onnx_reader/ie_onnx_reader.cpp
@@ -0,0 +1,36 @@
+// Copyright (C) 2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include "ie_onnx_reader.hpp"
+#include <ie_api.h>
+#include <ngraph/frontend/onnx_import/onnx.hpp>
+
+using namespace InferenceEngine;
+
+bool ONNXReader::supportModel(std::istream& model) const {
+    model.seekg(0, model.beg);
+    const int header_size = 128;
+    std::string header(header_size, ' ');
+    model.read(&header[0], header_size);
+    // find 'onnx' substring in the .onnx files
+    // find 'ir_version' and 'graph' for prototxt
+    // return (header.find("onnx") != std::string::npos) || (header.find("pytorch") != std::string::npos) ||
+    //     (header.find("ir_version") != std::string::npos && header.find("graph") != std::string::npos);
+    return !((header.find("<net ") != std::string::npos) || (header.find("<Net ") != std::string::npos));
+}
+
+CNNNetwork ONNXReader::read(std::istream& model, const std::vector<IExtensionPtr>& exts) const {
+    model.seekg(0, model.beg);
+    return CNNNetwork(ngraph::onnx_import::import_onnx_model(model));
+}
+
+INFERENCE_PLUGIN_API(StatusCode) InferenceEngine::CreateReader(IReader*& reader, ResponseDesc *resp) noexcept {
+    try {
+        reader = new ONNXReader();
+        return OK;
+    }
+    catch (std::exception &ex) {
+        return GENERAL_ERROR;
+    }
+}
diff --git a/inference-engine/src/readers/onnx_reader/ie_onnx_reader.hpp b/inference-engine/src/readers/onnx_reader/ie_onnx_reader.hpp
new file mode 100644
index 00000000000..b5335735f2d
--- /dev/null
+++ b/inference-engine/src/readers/onnx_reader/ie_onnx_reader.hpp
@@ -0,0 +1,48 @@
+// Copyright (C) 2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include <ie_reader.hpp>
+
+namespace InferenceEngine {
+
+class ONNXReader: public IReader {
+public:
+    void Release() noexcept override {
+        delete this;
+    }
+    /**
+     * @brief Checks that reader supports format of the model
+     * @param model stream with model
+     * @return true if format is supported
+     */
+    bool supportModel(std::istream& model) const override;
+    /**
+     * @brief Reads the model to CNNNetwork
+     * @param model stream with model
+     * @param exts vector with extensions
+     *
+     * @return CNNNetwork
+     */
+    CNNNetwork read(std::istream& model, const std::vector<IExtensionPtr>& exts) const override;
+    /**
+     * @brief Reads the model to CNNNetwork
+     * @param model stream with model
+     * @param weights stream with binary data
+     * @param exts vector with extensions
+     *
+     * @return CNNNetwork
+     */
+    CNNNetwork read(std::istream& model, std::istream& weights, const std::vector<IExtensionPtr>& exts) const override {
+        THROW_IE_EXCEPTION << "ONNX reader cannot read model with weights!";
+    }
+
+    std::vector<std::string> getDataFileExtensions() const override {
+        return {};
+    }
+};
+
+}  // namespace InferenceEngine
+
diff --git a/inference-engine/src/readers/reader_api/ie_blob_stream.hpp b/inference-engine/src/readers/reader_api/ie_blob_stream.hpp
new file mode 100644
index 00000000000..f4a3e5ec957
--- /dev/null
+++ b/inference-engine/src/readers/reader_api/ie_blob_stream.hpp
@@ -0,0 +1,35 @@
+// Copyright (C) 2018-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include <ie_blob.h>
+#include <istream>
+
+namespace InferenceEngine {
+namespace details {
+
+class INFERENCE_ENGINE_API_CLASS(BlobStream): public std::istream {
+private:
+    class BlobBuffer: public std::streambuf {
+    public:
+        BlobBuffer(const Blob::CPtr& blob);
+        ~BlobBuffer() override;
+        std::streampos seekpos(std::streampos sp, std::ios_base::openmode which) override;
+        std::streampos seekoff(std::streamoff off, std::ios_base::seekdir way, std::ios_base::openmode which) override;
+    };
+
+    BlobBuffer buffer;
+    Blob::CPtr blob;
+
+public:
+    BlobStream(const Blob::CPtr& blob);
+    ~BlobStream() override;
+
+    Blob::CPtr getBlob();
+};
+
+
+}  // namespace details
+}  // namespace InferenceEngine
diff --git a/inference-engine/src/readers/reader_api/ie_reader.hpp b/inference-engine/src/readers/reader_api/ie_reader.hpp
new file mode 100644
index 00000000000..d8e51f39916
--- /dev/null
+++ b/inference-engine/src/readers/reader_api/ie_reader.hpp
@@ -0,0 +1,63 @@
+// Copyright (C) 2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include <details/ie_irelease.hpp>
+#include <cpp/ie_cnn_network.h>
+#include <ie_iextension.h>
+#include <istream>
+#include <string>
+#include <vector>
+#include <ie_api.h>
+
+namespace InferenceEngine {
+
+/**
+ * @brief IReader an abstract interface for Inference Engine readers
+ */
+class INFERENCE_ENGINE_API_CLASS(IReader): public details::IRelease {
+public:
+    /**
+     * @brief Checks that reader supports format of the model
+     * @param model stream with model
+     * @return true if format is supported
+     */
+    virtual bool supportModel(std::istream& model) const = 0;
+    /**
+     * @brief Reads the model to CNNNetwork
+     * @param model stream with model
+     * @param exts vector with extensions
+     *
+     * @return CNNNetwork
+     */
+    virtual CNNNetwork read(std::istream& model, const std::vector<IExtensionPtr>& exts) const = 0;
+    /**
+     * @brief Reads the model to CNNNetwork
+     * @param model stream with model
+     * @param weights stream with binary data
+     * @param exts vector with extensions
+     *
+     * @return CNNNetwork
+     */
+    virtual CNNNetwork read(std::istream& model, std::istream& weights, const std::vector<IExtensionPtr>& exts) const = 0;
+
+    /**
+     * @brief Returns all supported extensions for data files
+     *
+     * @return vector of file extensions, for example the reader for OpenVINO IR returns {"bin"}
+     */
+    virtual std::vector<std::string> getDataFileExtensions() const = 0;
+};
+
+/**
+ * @brief Creates the default instance of the reader
+ *
+ * @param reader Reader interface
+ * @param resp Response description
+ * @return Status code
+ */
+INFERENCE_PLUGIN_API(StatusCode) CreateReader(IReader*& reader, ResponseDesc* resp) noexcept;
+
+}  // namespace InferenceEngine
diff --git a/inference-engine/src/readers/reader_api/ie_reader_ptr.hpp b/inference-engine/src/readers/reader_api/ie_reader_ptr.hpp
new file mode 100644
index 00000000000..9c3aee3ac51
--- /dev/null
+++ b/inference-engine/src/readers/reader_api/ie_reader_ptr.hpp
@@ -0,0 +1,36 @@
+// Copyright (C) 2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#pragma once
+
+#include <string>
+
+#include <details/ie_so_pointer.hpp>
+#include "ie_reader.hpp"
+
+namespace InferenceEngine {
+namespace details {
+
+/**
+ * @brief This class defines the name of the fabric for creating an IReader object in DLL
+ */
+template <>
+class SOCreatorTrait<IReader> {
+public:
+    /**
+     * @brief A name of the fabric for creating IReader object in DLL
+     */
+    static constexpr auto name = "CreateReader";
+};
+
+}  // namespace details
+
+/**
+ * @brief A C++ helper to work with objects created by the plugin.
+ *
+ * Implements different interfaces.
+ */
+using IReaderPtr = InferenceEngine::details::SOPointer<IReader>;
+
+}  // namespace InferenceEngine
diff --git a/inference-engine/tests/functional/inference_engine/local_test.cpp b/inference-engine/tests/functional/inference_engine/local_test.cpp
index 73af0064f02..54bbc72c9b9 100644
--- a/inference-engine/tests/functional/inference_engine/local_test.cpp
+++ b/inference-engine/tests/functional/inference_engine/local_test.cpp
@@ -207,7 +207,9 @@ protected:
         // This model contains layers with float attributes.
         // Conversion from string may be affected by locale.
         std::string model = isLSTM ? _model_LSTM : _model;
-        auto net = core.ReadNetwork(model, InferenceEngine::Blob::CPtr());
+        auto blob = make_shared_blob<uint8_t>(TensorDesc(Precision::U8, {3360}, Layout::C));
+        blob->allocate();
+        auto net = core.ReadNetwork(model, blob);
 
         IE_SUPPRESS_DEPRECATED_START
         if (!isLSTM) {
diff --git a/inference-engine/tests/functional/inference_engine/net_reader_test.cpp b/inference-engine/tests/functional/inference_engine/net_reader_test.cpp
index 2d008bbd905..4df6db84f56 100644
--- a/inference-engine/tests/functional/inference_engine/net_reader_test.cpp
+++ b/inference-engine/tests/functional/inference_engine/net_reader_test.cpp
@@ -107,7 +107,7 @@ TEST_P(NetReaderTest, ReadNetworkTwiceSeparately) {
 
 #ifdef ENABLE_UNICODE_PATH_SUPPORT
 
-TEST_P(NetReaderTest, ReadCorrectModelWithWeightsUnicodePath) {
+TEST_P(NetReaderTest, DISABLED_ReadCorrectModelWithWeightsUnicodePath) {
     GTEST_COUT << "params.modelPath: '" << _modelPath << "'" << std::endl;
     GTEST_COUT << "params.weightsPath: '" << _weightsPath << "'" << std::endl;
     GTEST_COUT << "params.netPrc: '" << _netPrc.name() << "'" << std::endl;
diff --git a/inference-engine/tests/functional/inference_engine/onnx_reader/onnx_reader.cpp b/inference-engine/tests/functional/inference_engine/onnx_reader/onnx_reader.cpp
new file mode 100644
index 00000000000..a74ed287783
--- /dev/null
+++ b/inference-engine/tests/functional/inference_engine/onnx_reader/onnx_reader.cpp
@@ -0,0 +1,136 @@
+// Copyright (C) 2018-2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include <gtest/gtest.h>
+#include <set>
+#include <string>
+#include <fstream>
+
+#include <ie_blob.h>
+#include <ie_core.hpp>
+#include <ngraph/ngraph.hpp>
+
+TEST(ONNX_Reader_Tests, ImportBasicModelToCore) {
+    std::string model = R"V0G0N(
+ir_version: 3
+producer_name: "nGraph ONNX Importer"
+graph {
+  node {
+    output: "B"
+    op_type: "Constant"
+    attribute {
+      name: "value"
+      t {
+        dims: 2
+        dims: 2
+        data_type: 1
+        float_data: 1
+        float_data: 2
+        float_data: 3
+        float_data: 4
+        name: "const_tensor"
+      }
+      type: TENSOR
+    }
+  }
+  node {
+    input: "A"
+    input: "B"
+    output: "X"
+    name: "add_node1"
+    op_type: "Add"
+  }
+  node {
+    input: "X"
+    input: "C"
+    output: "Y"
+    name: "add_node2"
+    op_type: "Add"
+  }
+  name: "test_graph"
+  initializer {
+    dims: 2
+    dims: 2
+    data_type: 1
+    name: "A"
+    raw_data: "\000\000\200?\000\000\000@\000\000@@\000\000\200@"
+  }
+  input {
+    name: "A"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+  input {
+    name: "C"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+  output {
+    name: "Y"
+    type {
+      tensor_type {
+        elem_type: 1
+        shape {
+          dim {
+            dim_value: 2
+          }
+          dim {
+            dim_value: 2
+          }
+        }
+      }
+    }
+  }
+}
+opset_import {
+  version: 4
+}
+)V0G0N";
+    InferenceEngine::Core ie;
+    InferenceEngine::Blob::CPtr weights;
+    auto cnnNetwork = ie.ReadNetwork(model, weights);
+    auto function = cnnNetwork.getFunction();
+
+    int count_additions = 0;
+    int count_constants = 0;
+    int count_parameters = 0;
+
+    for (auto op : function->get_ops()) {
+        const auto op_type = std::string(op->get_type_name());
+        count_additions += (op_type == "Add" ? 1 : 0);
+        count_constants += (op_type == "Constant" ? 1 : 0);
+        count_parameters += (op_type == "Parameter" ? 1 : 0);
+    }
+
+    ASSERT_EQ(function->get_output_size(), 1);
+    ASSERT_EQ(std::string(function->get_output_op(0)->get_type_name()), "Result");
+    ASSERT_EQ(function->get_output_element_type(0), ngraph::element::f32);
+    ASSERT_EQ(function->get_output_shape(0), ngraph::Shape({2, 2}));
+    ASSERT_EQ(count_additions, 2);
+    ASSERT_EQ(count_constants, 2);
+    ASSERT_EQ(count_parameters, 1);
+}
+
diff --git a/inference-engine/tests/ie_test_utils/common_test_utils/test_assertions.hpp b/inference-engine/tests/ie_test_utils/common_test_utils/test_assertions.hpp
index b74ac0aaef0..d737e69fea6 100644
--- a/inference-engine/tests/ie_test_utils/common_test_utils/test_assertions.hpp
+++ b/inference-engine/tests/ie_test_utils/common_test_utils/test_assertions.hpp
@@ -85,7 +85,7 @@ inline void compare_preprocess(const InferenceEngine::PreProcessChannel & lhs, c
 inline void compare_preprocess_info(const InferenceEngine::PreProcessInfo & lhs, const InferenceEngine::PreProcessInfo & rhs) {
     ASSERT_EQ(lhs.getMeanVariant(), rhs.getMeanVariant());
     ASSERT_EQ(lhs.getNumberOfChannels(), rhs.getNumberOfChannels());
-    for (int i = 0; i < lhs.getNumberOfChannels(); i++) {
+    for (size_t i = 0; i < lhs.getNumberOfChannels(); i++) {
         ASSERT_PREPROCESS_CHANNEL_EQ(*lhs[i].get(), *rhs[i].get());
     }
 }
@@ -95,7 +95,7 @@ inline void compare_outputs_info(const InferenceEngine::OutputsDataMap & lhs, co
     auto i = lhs.begin();
     auto j = rhs.begin();
 
-    for (int k =0; k != lhs.size(); k++, i++, j++) {
+    for (size_t k =0; k != lhs.size(); k++, i++, j++) {
         ASSERT_STREQ(i->first.c_str(), j->first.c_str());
         ASSERT_DATA_EQ(*i->second.get(), *j->second.get());
     }
@@ -106,7 +106,7 @@ inline void compare_inputs_info(const InferenceEngine::InputsDataMap & lhs, cons
     auto i = lhs.begin();
     auto j = rhs.begin();
 
-    for (int k = 0; k != lhs.size(); k++, i++, j++) {
+    for (size_t k = 0; k != lhs.size(); k++, i++, j++) {
         ASSERT_STREQ(i->first.c_str(), j->first.c_str());
         ASSERT_DIMS_EQ(i->second->getTensorDesc().getDims(), j->second->getTensorDesc().getDims());
         ASSERT_PREPROCESS_INFO_EQ(i->second->getPreProcess(), j->second->getPreProcess());
diff --git a/inference-engine/tests/ie_test_utils/functional_test_utils/network_utils.cpp b/inference-engine/tests/ie_test_utils/functional_test_utils/network_utils.cpp
index e5ed6896757..aeae4ed00cc 100644
--- a/inference-engine/tests/ie_test_utils/functional_test_utils/network_utils.cpp
+++ b/inference-engine/tests/ie_test_utils/functional_test_utils/network_utils.cpp
@@ -264,9 +264,6 @@ namespace FuncTestUtils {
         auto back_edges_mp_old = get_port_map(ti_old->back_edges, ti_old->body.outputs, ti_old->body.inputs);
         compare_port_maps(back_edges_mp_new, back_edges_mp_old);
 
-        auto holder = ti_new->body.inputs.back();
-        ti_new->body.inputs.pop_back();
-
         // TI body comparison
         auto nodes_new = InferenceEngine::NetPass::TIBodySortTopologically(ti_new->body);
         auto nodes_old = InferenceEngine::NetPass::TIBodySortTopologically(ti_old->body);
@@ -303,8 +300,6 @@ namespace FuncTestUtils {
         auto old_outputs = get_map(ti_old->body.outputs);
         compareInfo<std::map<std::string, InferenceEngine::DataPtr>>(new_outputs, old_outputs,
                                                                      "Bodies of TensorIterator have different outputs!");
-
-        ti_new->body.inputs.push_back(holder);
         IE_SUPPRESS_DEPRECATED_END
     }
 
diff --git a/inference-engine/tests/ie_test_utils/unit_test_utils/CMakeLists.txt b/inference-engine/tests/ie_test_utils/unit_test_utils/CMakeLists.txt
index c6cafb304ff..91076ff09ff 100644
--- a/inference-engine/tests/ie_test_utils/unit_test_utils/CMakeLists.txt
+++ b/inference-engine/tests/ie_test_utils/unit_test_utils/CMakeLists.txt
@@ -8,7 +8,7 @@ list(APPEND EXPORT_DEPENDENCIES
         commonTestUtils_s
         inference_engine_s
         inference_engine_lp_transformations
-        inference_engine_ir_readers
+        inference_engine_ir_reader
         gmock)
 
 addIeTarget(
@@ -23,4 +23,4 @@ addIeTarget(
 
 target_link_libraries(${TARGET_NAME}
         PUBLIC
-        ${EXPORT_DEPENDENCIES})
\ No newline at end of file
+        ${EXPORT_DEPENDENCIES})
diff --git a/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_infer_request.hpp b/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_infer_request.hpp
index b0fea6154d5..503e6fabfb0 100644
--- a/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_infer_request.hpp
+++ b/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_infer_request.hpp
@@ -42,10 +42,7 @@ TEST_P(BehaviorPluginTestInferRequest, SetEmptyConfig) {
 // Load incorrect network to Plugin to get executable network
 TEST_P(BehaviorPluginTestInferRequest, canNotLoadNetworkToGetExeNetworkWithoutWeights) {
     InferenceEngine::Core core;
-    CNNNetwork network = core.ReadNetwork(GetParam().model_xml_str, Blob::CPtr());
-
-    ASSERT_THROW(core.LoadNetwork(network, GetParam().device, GetParam().config),
-                 InferenceEngineException);
+    ASSERT_THROW(core.ReadNetwork(GetParam().model_xml_str, Blob::CPtr()), InferenceEngineException);
 }
 
 // Load correct network to Plugin to get executable network
diff --git a/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_layers.hpp b/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_layers.hpp
index 396789867f1..e90a45906a5 100644
--- a/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_layers.hpp
+++ b/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugin_layers.hpp
@@ -349,7 +349,7 @@ protected:
         V2NetBuilder model = V2NetBuilder::buildNetworkWithOneInput(
                 "Activation_Only", inout.inDims[0], p.precision)
                 .addLayer("Activation", p.precision, &params, inout);
-        return model.finish();
+        return model.finish(false);
     }
 };
 
diff --git a/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugins.hpp b/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugins.hpp
index fe63d50acd7..5fb51ea0c0f 100644
--- a/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugins.hpp
+++ b/inference-engine/tests_deprecated/behavior/shared_tests/plugin_tests/behavior_test_plugins.hpp
@@ -96,10 +96,7 @@ TEST_F(BehaviorPluginTest, smoke_llocateNullBlob) {
 // Load incorrect network to Plugin
 TEST_P(BehaviorPluginTest, canNotLoadNetworkWithoutWeights) {
     InferenceEngine::Core core;
-    CNNNetwork network = core.ReadNetwork(GetParam().model_xml_str, Blob::CPtr());
-
-    IExecutableNetwork::Ptr exeNetwork;
-    ASSERT_THROW(core.LoadNetwork(network, GetParam().device, {}), InferenceEngineException);
+    ASSERT_THROW(core.ReadNetwork(GetParam().model_xml_str, Blob::CPtr()), InferenceEngineException);
 }
 
 bool static compare_two_files_lexicographically(const std::string& name_a, const std::string& name_b) {
diff --git a/inference-engine/tests_deprecated/behavior/vpu/myriad_tests/vpu_watchdog_tests.cpp b/inference-engine/tests_deprecated/behavior/vpu/myriad_tests/vpu_watchdog_tests.cpp
index 8340d2a9065..a1f63edafb4 100644
--- a/inference-engine/tests_deprecated/behavior/vpu/myriad_tests/vpu_watchdog_tests.cpp
+++ b/inference-engine/tests_deprecated/behavior/vpu/myriad_tests/vpu_watchdog_tests.cpp
@@ -181,7 +181,12 @@ TEST_P(MYRIADWatchdog, watchDogIntervalDefault) {
     {
 
         InferenceEngine::Core core;
-        CNNNetwork network = core.ReadNetwork(GetParam().model_xml_str, Blob::CPtr());
+        IE_SUPPRESS_DEPRECATED_START
+        CNNNetReader reader;
+        reader.ReadNetwork(GetParam().model_xml_str.data(), GetParam().model_xml_str.length());
+
+        CNNNetwork network = reader.getNetwork();
+        IE_SUPPRESS_DEPRECATED_END
         ASSERT_GE(startup_devices.unbooted, 1);
 
         ExecutableNetwork ret;
@@ -214,7 +219,12 @@ TEST_P(MYRIADWatchdog, canTurnoffWatchDogViaConfig) {
     auto ctime = Time::now();
     {
         InferenceEngine::Core core;
-        CNNNetwork network = core.ReadNetwork(GetParam().model_xml_str, Blob::CPtr());
+        IE_SUPPRESS_DEPRECATED_START
+        CNNNetReader reader;
+        reader.ReadNetwork(GetParam().model_xml_str.data(), GetParam().model_xml_str.length());
+
+        CNNNetwork network = reader.getNetwork();
+        IE_SUPPRESS_DEPRECATED_END
         ASSERT_GE(startup_devices.unbooted, 1);
 
         ExecutableNetwork ret;
diff --git a/inference-engine/tests_deprecated/functional/mkldnn/network_tests/ngraph_network_test.cpp b/inference-engine/tests_deprecated/functional/mkldnn/network_tests/ngraph_network_test.cpp
index fd08dd121aa..b645bfa11bf 100644
--- a/inference-engine/tests_deprecated/functional/mkldnn/network_tests/ngraph_network_test.cpp
+++ b/inference-engine/tests_deprecated/functional/mkldnn/network_tests/ngraph_network_test.cpp
@@ -7,7 +7,6 @@
 #include <tests_common_func.hpp>
 #include <memory>
 #include "xml_helper.hpp"
-#include <ie_ir_reader.hpp>
 #include <ie_core.hpp>
 
 #define XBYAK_NO_OP_NAMES
@@ -54,12 +53,9 @@ struct ngraph_network_param {
 class smoke_NGraphNetworkTest : public TestsCommon, public TestsCommonFunc {
 protected:
     Blob::Ptr classifyV7(ngraph_network_param p, size_t batch_size = 1, float threshold = 0.005f) {
-        IRReader reader;
-        auto ngraph = reader.read(p.v7model());
-
-        auto network = CNNNetwork(ngraph);
-
         Core ie;
+        CNNNetwork network = ie.ReadNetwork(p.v7model());
+
         ExecutableNetwork exeNetwork = ie.LoadNetwork(network, "CPU");
         InferRequest inferRequest = exeNetwork.CreateInferRequest();
 
diff --git a/inference-engine/tests_deprecated/helpers/CMakeLists.txt b/inference-engine/tests_deprecated/helpers/CMakeLists.txt
index 1777ac9071c..bf915e52d1a 100644
--- a/inference-engine/tests_deprecated/helpers/CMakeLists.txt
+++ b/inference-engine/tests_deprecated/helpers/CMakeLists.txt
@@ -21,7 +21,7 @@ function(add_helpers target_name)
 
     target_include_directories(${target_name} PUBLIC "${CMAKE_CURRENT_SOURCE_DIR}"
                                                      "${IE_MAIN_SOURCE_DIR}/src/inference_engine"
-                                                     $<TARGET_PROPERTY:inference_engine_ir_readers,INTERFACE_INCLUDE_DIRECTORIES>
+                                                     $<TARGET_PROPERTY:inference_engine_ir_reader,INTERFACE_INCLUDE_DIRECTORIES>
                                                      $<TARGET_PROPERTY:inference_engine_lp_transformations,INTERFACE_INCLUDE_DIRECTORIES>
                                                      $<TARGET_PROPERTY:pugixml,INTERFACE_INCLUDE_DIRECTORIES>
                                                      "${IE_MAIN_SOURCE_DIR}/src/vpu/"
diff --git a/inference-engine/tests_deprecated/unit/CMakeLists.txt b/inference-engine/tests_deprecated/unit/CMakeLists.txt
index 9f614a276c8..48afcbb7016 100644
--- a/inference-engine/tests_deprecated/unit/CMakeLists.txt
+++ b/inference-engine/tests_deprecated/unit/CMakeLists.txt
@@ -148,8 +148,8 @@ target_link_libraries(${TARGET_NAME} PRIVATE
 
     # dynamic libraries
     inference_engine_lp_transformations
-    inference_engine_ir_readers
     inference_engine_transformations
+    inference_engine_ir_reader
     ${CMAKE_DL_LIBS})
 
 if(TARGET libGNAStubs)
diff --git a/inference-engine/tests_deprecated/unit/cnn_network/cnn_net_reader_impl_test.cpp b/inference-engine/tests_deprecated/unit/cnn_network/cnn_net_reader_impl_test.cpp
index 7c940a6fe06..e8ff39380f6 100644
--- a/inference-engine/tests_deprecated/unit/cnn_network/cnn_net_reader_impl_test.cpp
+++ b/inference-engine/tests_deprecated/unit/cnn_network/cnn_net_reader_impl_test.cpp
@@ -2291,7 +2291,9 @@ TEST_F(CNNNetReaderImplTest, ReadInThreads) {
         threads.push_back(std::thread([i, model]{
                     InferenceEngine::Core core;
                     /** Read network model **/
-                    CNNNetwork network = core.ReadNetwork(model, InferenceEngine::Blob::CPtr());
+                    auto blob = make_shared_blob<uint8_t>(TensorDesc(Precision::U8, {9728}, Layout::C));
+                    blob->allocate();
+                    CNNNetwork network = core.ReadNetwork(model, blob);
                     // -----------------------------------------------------------------------------------------------------
 
                     // --------------------------- 3. Configure input & output ---------------------------------------------
diff --git a/inference-engine/tests_deprecated/unit/engines/mkldnn/graph/layers/internal/graph_concat_test.cpp b/inference-engine/tests_deprecated/unit/engines/mkldnn/graph/layers/internal/graph_concat_test.cpp
index d44cc2d0dd6..01f80bad5c1 100644
--- a/inference-engine/tests_deprecated/unit/engines/mkldnn/graph/layers/internal/graph_concat_test.cpp
+++ b/inference-engine/tests_deprecated/unit/engines/mkldnn/graph/layers/internal/graph_concat_test.cpp
@@ -1102,8 +1102,9 @@ protected:
             std::string model = getModel(p);
 
             InferenceEngine::Core core;
-            ASSERT_THROW(core.ReadNetwork(model, InferenceEngine::Blob::CPtr()), 
-                         InferenceEngine::details::InferenceEngineException);
+            // TODO: check InferenceEngine::details::InferenceEngineException when RTTI issue will be resolved
+            ASSERT_THROW(core.ReadNetwork(model, InferenceEngine::Blob::CPtr()),
+                         std::exception);
         } catch (const InferenceEngine::details::InferenceEngineException &e) {
             FAIL() << e.what();
         }
diff --git a/inference-engine/tests_deprecated/unit/graph_tools/graph_copy_tests.cpp b/inference-engine/tests_deprecated/unit/graph_tools/graph_copy_tests.cpp
index 0541458b06d..9895176e47c 100644
--- a/inference-engine/tests_deprecated/unit/graph_tools/graph_copy_tests.cpp
+++ b/inference-engine/tests_deprecated/unit/graph_tools/graph_copy_tests.cpp
@@ -278,6 +278,7 @@ TEST(CNNSpecificGraphCopyTests, copyPreprocess) {
 
     InferenceEngine::Core core;
     InferenceEngine::CNNNetwork network;
+
     ASSERT_NO_THROW(network = core.ReadNetwork(SINGLE_LAYER_MODEL, InferenceEngine::Blob::CPtr()));
 
     //copy the network
@@ -330,7 +331,7 @@ TEST(CNNSpecificGraphCopyTests, copyNetworkWithDeconvolution) {
                     <dim>4</dim>
                 </port>
             </output>
-            <weights offset="5517824" size="12288"/>
+            <weights offset="0" size="12288"/>
         </layer>
         </layers>
         <edges>
@@ -341,7 +342,9 @@ TEST(CNNSpecificGraphCopyTests, copyNetworkWithDeconvolution) {
 
     InferenceEngine::Core core;
     InferenceEngine::CNNNetwork network;
-    ASSERT_NO_THROW(network = core.ReadNetwork(SINGLE_LAYER_MODEL, InferenceEngine::Blob::CPtr()));
+    auto blob = make_shared_blob<uint8_t>(TensorDesc(Precision::U8, {12288}, Layout::C));
+    blob->allocate();
+    ASSERT_NO_THROW(network = core.ReadNetwork(SINGLE_LAYER_MODEL, blob));
 
     // copy the network
     struct EmptyStruct {};
diff --git a/inference-engine/tests_deprecated/unit/inference_engine_tests/normalization/latest_in_fuse_test.cpp b/inference-engine/tests_deprecated/unit/inference_engine_tests/normalization/latest_in_fuse_test.cpp
index ea45be329c0..bfea14be947 100644
--- a/inference-engine/tests_deprecated/unit/inference_engine_tests/normalization/latest_in_fuse_test.cpp
+++ b/inference-engine/tests_deprecated/unit/inference_engine_tests/normalization/latest_in_fuse_test.cpp
@@ -87,7 +87,7 @@ class NormalizationConvConvEltwiseTests: public TestsCommon,
 
     std::string getModel(conv_conv_eltwise_params p) {
         std::string model = layers_t;
-        
+
         std::string s_dims;
         for (auto& dim : p.in) {
             s_dims += "\n                    <dim>";
@@ -132,7 +132,10 @@ protected:
 
             InferenceEngine::Core ie;
             InferenceEngine::CNNNetwork network;
-            ASSERT_NO_THROW(network = ie.ReadNetwork(model, InferenceEngine::Blob::CPtr()));
+            auto blob = InferenceEngine::make_shared_blob<uint8_t>(InferenceEngine::TensorDesc(InferenceEngine::Precision::U8,
+                                                                                               {9}, InferenceEngine::Layout::C));
+            blob->allocate();
+            ASSERT_NO_THROW(network = ie.ReadNetwork(model, blob));
 
             int maxSign = 0x7F;
             int maxUnsign = 0xFF;
@@ -156,7 +159,7 @@ TEST_P(NormalizationConvConvEltwiseTests, TestsConvConvEltwise) {}
 INSTANTIATE_TEST_CASE_P(
         TestsConvConvEltwise, NormalizationConvConvEltwiseTests,
         ::testing::Values(
-                conv_conv_eltwise_params{{1, 16, 4, 4}, 
+                conv_conv_eltwise_params{{1, 16, 4, 4},
                                      { {1, 1}, {1, 1}, {0, 0}, {0, 0}, {1, 1}, "", 1, 32, true },
                                      {"sum", {}} },
                 conv_conv_eltwise_params{{1, 16, 4, 4, 4},