diff --git a/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/border/border_kernel_ref.cpp b/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/border/border_kernel_ref.cpp
index bed25eb288a..a0b2aecb98c 100644
--- a/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/border/border_kernel_ref.cpp
+++ b/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/border/border_kernel_ref.cpp
@@ -24,6 +24,7 @@ ParamsKey BorderKernelRef::GetSupportedKey() const {
     k.EnableInputLayout(DataLayout::bfzyx);
     k.EnableInputLayout(DataLayout::bfwzyx);
     k.EnableInputLayout(DataLayout::b_fs_yx_fsv16);
+    k.EnableInputLayout(DataLayout::b_fs_yx_fsv32);
     k.EnableInputLayout(DataLayout::b_fs_zyx_fsv16);
 
     k.EnableOutputLayout(DataLayout::bfyx);
@@ -32,6 +33,7 @@ ParamsKey BorderKernelRef::GetSupportedKey() const {
     k.EnableOutputLayout(DataLayout::bfzyx);
     k.EnableOutputLayout(DataLayout::bfwzyx);
     k.EnableOutputLayout(DataLayout::b_fs_yx_fsv16);
+    k.EnableOutputLayout(DataLayout::b_fs_yx_fsv32);
     k.EnableOutputLayout(DataLayout::b_fs_zyx_fsv16);
 
     k.EnableTensorOffset();
diff --git a/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/convolution/convolution_kernel_b_fs_zyx_fsv16_imad.cpp b/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/convolution/convolution_kernel_b_fs_zyx_fsv16_imad.cpp
index 87f09fd2606..a7a971c38fb 100644
--- a/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/convolution/convolution_kernel_b_fs_zyx_fsv16_imad.cpp
+++ b/inference-engine/thirdparty/clDNN/kernel_selector/core/actual_kernels/convolution/convolution_kernel_b_fs_zyx_fsv16_imad.cpp
@@ -78,6 +78,13 @@ Convolution_kernel_b_fs_zyx_fsv16_imad::GetBlockParams(const convolution_params&
 
     // Estimate basic block params ratio
     auto test_block_params = BlockParams{ block_width, 1, 1, simd, in_block_width, 1, 1, 1 };
+
+    // Use default block parameters for asymmetric weights quantization for devices with immad support due to unoptimized tuning
+    if ((params.quantization == QuantizationType::ASYMMETRIC_DATA_AND_WEIGHTS || params.quantization == QuantizationType::ASYMMETRIC_WEIGHTS) &&
+        params.engineInfo.bIMMADSupport) {
+        return test_block_params;
+    }
+
     auto best_block_params_ratio = EstimateBlockParamsRatio(params, test_block_params);
 
     size_t max_slm_split = params.engineInfo.maxWorkGroupSize / simd;
diff --git a/inference-engine/thirdparty/clDNN/src/impls/ocl/border.cpp b/inference-engine/thirdparty/clDNN/src/impls/ocl/border.cpp
index 02a3d1657f5..148579ba485 100644
--- a/inference-engine/thirdparty/clDNN/src/impls/ocl/border.cpp
+++ b/inference-engine/thirdparty/clDNN/src/impls/ocl/border.cpp
@@ -98,6 +98,11 @@ attach_border_impl::attach_border_impl() {
         std::make_tuple(data_types::i8, format::b_fs_yx_fsv16),
         std::make_tuple(data_types::u8, format::b_fs_yx_fsv16),
 
+        std::make_tuple(data_types::f32, format::b_fs_yx_fsv32),
+        std::make_tuple(data_types::f16, format::b_fs_yx_fsv32),
+        std::make_tuple(data_types::i8, format::b_fs_yx_fsv32),
+        std::make_tuple(data_types::u8, format::b_fs_yx_fsv32),
+
         std::make_tuple(data_types::f32, format::b_fs_zyx_fsv16),
         std::make_tuple(data_types::f16, format::b_fs_zyx_fsv16),
         std::make_tuple(data_types::i8, format::b_fs_zyx_fsv16),