diff --git a/src/core/CL/kernels/CLBatchConcatenateLayerKernel.cpp b/src/core/CL/kernels/CLBatchConcatenateLayerKernel.cpp
index 9c2b68a..b3496f7 100644
--- a/src/core/CL/kernels/CLBatchConcatenateLayerKernel.cpp
+++ b/src/core/CL/kernels/CLBatchConcatenateLayerKernel.cpp
@@ -28,9 +28,9 @@
 #include "arm_compute/core/CL/CLValidate.h"
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/Cast.h"
 
 #include "support/StringSupport.h"
 
@@ -75,34 +75,27 @@
 } // namespace
 
 CLBatchConcatenateLayerKernel::CLBatchConcatenateLayerKernel()
-    : _input(nullptr), _output(nullptr), _batch_offset(0)
+    : _batch_offset(0)
 {
 }
 
-void CLBatchConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int batch_offset, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, batch_offset, output);
-}
-
-void CLBatchConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int batch_offset, ICLTensor *output)
+void CLBatchConcatenateLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, unsigned int batch_offset, ITensorInfo *output)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), batch_offset, output->info()));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, batch_offset, output));
 
-    _input        = input;
-    _output       = output;
     _batch_offset = batch_offset;
 
-    const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size();
+    const unsigned int num_elems_processed_per_iteration = 16 / input->element_size();
 
     // Add build options
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->data_type()));
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
-    if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info())
+    if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info())
     {
-        const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
-        const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform();
+        const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
+        const UniformQuantizationInfo oq_info = output->quantization_info().uniform();
 
         build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset));
         build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
@@ -114,13 +107,13 @@
     _kernel = create_kernel(compile_context, "concatenate", build_opts.options());
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), batch_offset, output->info());
+    auto win_config = validate_and_configure_window(input, batch_offset, output);
     ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
 
     ICLKernel::configure_internal(std::get<1>(win_config));
 
     // Set output valid region
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
 
     // Set config_id for enabling LWS tuning
     _config_id = "concatenate_";
@@ -128,13 +121,13 @@
     _config_id += "_";
     _config_id += support::cpp11::to_string(batch_offset);
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input->dimension(1));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input->info()->dimension(2));
+    _config_id += support::cpp11::to_string(input->dimension(2));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input->info()->dimension(3));
+    _config_id += support::cpp11::to_string(input->dimension(3));
 }
 
 Status CLBatchConcatenateLayerKernel::validate(const arm_compute::ITensorInfo *input,
@@ -146,14 +139,18 @@
     return Status{};
 }
 
-void CLBatchConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLBatchConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                                           const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
     Window slice = window.first_slice_window_3D();
 
-    const int offset_to_first_elements_in_bytes = _batch_offset * _output->info()->strides_in_bytes()[3];
+    const int offset_to_first_elements_in_bytes = _batch_offset * dst->info()->strides_in_bytes()[3];
 
     unsigned int idx = 2 * num_arguments_per_3D_tensor(); // Skip the input and output parameters
     _kernel.setArg<cl_int>(idx, offset_to_first_elements_in_bytes);
@@ -161,8 +158,8 @@
     do
     {
         unsigned int idx = 0;
-        add_3D_tensor_argument(idx, _input, slice);
-        add_3D_tensor_argument(idx, _output, slice);
+        add_3D_tensor_argument(idx, src, slice);
+        add_3D_tensor_argument(idx, dst, slice);
         enqueue(queue, *this, slice, lws_hint());
     }
     while(window.slide_window_slice_3D(slice));
diff --git a/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp b/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp
index 10aa615..0bae901 100644
--- a/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp
+++ b/src/core/CL/kernels/CLDepthConcatenateLayerKernel.cpp
@@ -28,9 +28,9 @@
 #include "arm_compute/core/CL/CLValidate.h"
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/Cast.h"
 
 #include "support/StringSupport.h"
 
@@ -73,34 +73,27 @@
 } // namespace
 
 CLDepthConcatenateLayerKernel::CLDepthConcatenateLayerKernel()
-    : _input(nullptr), _output(nullptr), _depth_offset(0)
+    : _depth_offset(0)
 {
 }
 
-void CLDepthConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int depth_offset, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, depth_offset, output);
-}
-
-void CLDepthConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int depth_offset, ICLTensor *output)
+void CLDepthConcatenateLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, unsigned int depth_offset, ITensorInfo *output)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), depth_offset, output->info()));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, depth_offset, output));
 
-    _input        = input;
-    _output       = output;
     _depth_offset = depth_offset;
 
-    const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size();
+    const unsigned int num_elems_processed_per_iteration = 16 / input->element_size();
 
     // Add build options
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->data_type()));
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
-    if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info())
+    if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info())
     {
-        const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
-        const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform();
+        const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
+        const UniformQuantizationInfo oq_info = output->quantization_info().uniform();
 
         build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset));
         build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
@@ -112,13 +105,13 @@
     _kernel = create_kernel(compile_context, "concatenate", build_opts.options());
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), depth_offset, output->info());
+    auto win_config = validate_and_configure_window(input, depth_offset, output);
     ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
 
     ICLKernel::configure_internal(std::get<1>(win_config));
 
     // Set output valid region
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
 }
 
 Status CLDepthConcatenateLayerKernel::validate(const arm_compute::ITensorInfo *input,
@@ -130,14 +123,18 @@
     return Status{};
 }
 
-void CLDepthConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLDepthConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                                           const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
     Window slice = window.first_slice_window_3D();
 
-    const int offset_to_first_elements_in_bytes = _depth_offset * _output->info()->strides_in_bytes()[2];
+    const int offset_to_first_elements_in_bytes = _depth_offset * dst->info()->strides_in_bytes()[2];
 
     unsigned int idx = 2 * num_arguments_per_3D_tensor(); // Skip the input and output parameters
     _kernel.setArg<cl_int>(idx, offset_to_first_elements_in_bytes);
@@ -145,8 +142,8 @@
     do
     {
         unsigned int idx = 0;
-        add_3D_tensor_argument(idx, _input, slice);
-        add_3D_tensor_argument(idx, _output, slice);
+        add_3D_tensor_argument(idx, src, slice);
+        add_3D_tensor_argument(idx, dst, slice);
         enqueue(queue, *this, slice, lws_hint());
     }
     while(window.slide_window_slice_3D(slice));
diff --git a/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp b/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp
index b3bae46..5ab270c 100644
--- a/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp
+++ b/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp
@@ -27,19 +27,14 @@
 #include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/CLValidate.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/Error.h"
 #include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/IAccessWindow.h"
-#include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/Cast.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 #include "support/StringSupport.h"
 
-#include <map>
-
 namespace arm_compute
 {
 namespace
@@ -77,7 +72,7 @@
 } // namespace
 
 CLHeightConcatenateLayerKernel::CLHeightConcatenateLayerKernel()
-    : _input(nullptr), _output(nullptr), _height_offset(0), _num_elems_processed_per_iteration()
+    : _height_offset(0), _num_elems_processed_per_iteration()
 {
 }
 
@@ -89,33 +84,26 @@
     return Status{};
 }
 
-void CLHeightConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int height_offset, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, height_offset, output);
-}
-
-void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int height_offset, ICLTensor *output)
+void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, unsigned int height_offset, ITensorInfo *output)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), height_offset, output->info()));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, height_offset, output));
 
-    _input         = input;
-    _output        = output;
     _height_offset = height_offset;
 
-    auto win_config = validate_and_configure_window(input->info(), output->info(), _num_elems_processed_per_iteration);
+    auto win_config = validate_and_configure_window(input, output, _num_elems_processed_per_iteration);
 
     // Add build options
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->info()->element_size()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->element_size()));
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(_num_elems_processed_per_iteration));
     build_opts.add_option("-DHEIGHT_OFFSET=" + support::cpp11::to_string(_height_offset));
-    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->info()->dimension(2)));
+    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->dimension(2)));
 
-    if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info())
+    if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info())
     {
-        const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
-        const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform();
+        const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
+        const UniformQuantizationInfo oq_info = output->quantization_info().uniform();
 
         build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset));
         build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
@@ -132,17 +120,21 @@
     ICLKernel::configure_internal(std::get<1>(win_config));
 
     // Set output valid region
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
 }
 
-void CLHeightConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLHeightConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                                            const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
     unsigned int idx = 0;
-    add_4D_tensor_argument(idx, _input, window);
-    add_4D_tensor_argument(idx, _output, window);
+    add_4D_tensor_argument(idx, src, window);
+    add_4D_tensor_argument(idx, dst, window);
     enqueue(queue, *this, window, lws_hint());
 }
 } // namespace arm_compute
diff --git a/src/core/CL/kernels/CLWidthConcatenate2TensorsKernel.cpp b/src/core/CL/kernels/CLWidthConcatenate2TensorsKernel.cpp
index c87290d..88b69c7 100644
--- a/src/core/CL/kernels/CLWidthConcatenate2TensorsKernel.cpp
+++ b/src/core/CL/kernels/CLWidthConcatenate2TensorsKernel.cpp
@@ -28,11 +28,10 @@
 #include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/CLValidate.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/IAccessWindow.h"
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
 #include "arm_compute/core/utils/helpers/tensor_info.h"
+#include "arm_compute/core/utils/misc/Cast.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 #include "support/StringSupport.h"
@@ -79,11 +78,6 @@
 }
 } // namespace
 
-CLWidthConcatenate2TensorsKernel::CLWidthConcatenate2TensorsKernel()
-    : _input1(nullptr), _input2(nullptr), _output(nullptr)
-{
-}
-
 Status CLWidthConcatenate2TensorsKernel::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
 {
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input1, input2, output));
@@ -91,35 +85,26 @@
     return Status{};
 }
 
-void CLWidthConcatenate2TensorsKernel::configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output);
-}
-
-void CLWidthConcatenate2TensorsKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output)
+void CLWidthConcatenate2TensorsKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input1->info(), input2->info(), output->info()));
-
-    _input1 = input1;
-    _input2 = input2;
-    _output = output;
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input1, input2, output));
 
     // Add build options
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input1->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input1->data_type()));
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
-    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input1->info()->dimension(2)));
-    build_opts.add_option("-DINPUT1_WIDTH=" + support::cpp11::to_string(input1->info()->dimension(0)));
-    build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(input1->info()->element_size()));
+    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input1->dimension(2)));
+    build_opts.add_option("-DINPUT1_WIDTH=" + support::cpp11::to_string(input1->dimension(0)));
+    build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(input1->element_size()));
 
     // If input have different quantization info set quantization parameters needed for the re-quantization process
-    const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(output->info(), input1->info(), input2->info());
-    if(is_data_type_quantized_asymmetric(input1->info()->data_type()) && have_different_qinfo)
+    const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(output, input1, input2);
+    if(is_data_type_quantized_asymmetric(input1->data_type()) && have_different_qinfo)
     {
-        const UniformQuantizationInfo iq1_info = input1->info()->quantization_info().uniform();
-        const UniformQuantizationInfo iq2_info = input2->info()->quantization_info().uniform();
-        const UniformQuantizationInfo oq_info  = output->info()->quantization_info().uniform();
+        const UniformQuantizationInfo iq1_info = input1->quantization_info().uniform();
+        const UniformQuantizationInfo iq2_info = input2->quantization_info().uniform();
+        const UniformQuantizationInfo oq_info  = output->quantization_info().uniform();
 
         build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq1_info.offset));
         build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
@@ -133,16 +118,16 @@
     _kernel = create_kernel(compile_context, "concatenate_width_x2", build_opts.options());
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input1->info(), input2->info(), output->info());
+    auto win_config = validate_and_configure_window(input1, input2, output);
     ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
 
     ICLKernel::configure_internal(std::get<1>(win_config));
 
     // Set output valid region
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
 
     // Pass paddings as arguments to the kernel
-    const unsigned int input1_width         = input1->info()->dimension(0);
+    const unsigned int input1_width         = input1->dimension(0);
     const unsigned int input1_right_padding = ceil_to_multiple(input1_width, num_elems_processed_per_iteration) - input1_width;
     const unsigned int input2_left_padding  = input1_width % num_elems_processed_per_iteration;
     unsigned int       idx0                 = 3 * num_arguments_per_4D_tensor();
@@ -151,30 +136,35 @@
 
     // Set config_id for enabling LWS tuning
     _config_id = "concatenate_width_x2_";
-    _config_id += lower_string(string_from_data_type(input1->info()->data_type()));
+    _config_id += lower_string(string_from_data_type(input1->data_type()));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input1->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input1->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input1->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input1->dimension(1));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input2->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input2->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input2->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input2->dimension(1));
 }
 
-void CLWidthConcatenate2TensorsKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLWidthConcatenate2TensorsKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                                              const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
     Window slice = window.first_slice_window_4D();
 
+    const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC_VEC));
+    const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC_VEC + 1));
+    auto       dst  = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
     do
     {
         unsigned int idx = 0;
-        add_4D_tensor_argument(idx, _input1, slice);
-        add_4D_tensor_argument(idx, _input2, slice);
-        add_4D_tensor_argument(idx, _output, slice);
+        add_4D_tensor_argument(idx, src0, slice);
+        add_4D_tensor_argument(idx, src1, slice);
+        add_4D_tensor_argument(idx, dst, slice);
         enqueue(queue, *this, window, lws_hint());
     }
     while(window.slide_window_slice_4D(slice));
diff --git a/src/core/CL/kernels/CLWidthConcatenate4TensorsKernel.cpp b/src/core/CL/kernels/CLWidthConcatenate4TensorsKernel.cpp
index 84744a6..e49bb1c 100644
--- a/src/core/CL/kernels/CLWidthConcatenate4TensorsKernel.cpp
+++ b/src/core/CL/kernels/CLWidthConcatenate4TensorsKernel.cpp
@@ -32,6 +32,7 @@
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
 #include "arm_compute/core/utils/helpers/tensor_info.h"
+#include "arm_compute/core/utils/misc/Cast.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 #include "support/StringSupport.h"
@@ -97,7 +98,6 @@
 } // namespace
 
 CLWidthConcatenate4TensorsKernel::CLWidthConcatenate4TensorsKernel()
-    : _input1(nullptr), _input2(nullptr), _input3(nullptr), _input4(nullptr), _output(nullptr)
 {
 }
 
@@ -108,42 +108,33 @@
     return Status{};
 }
 
-void CLWidthConcatenate4TensorsKernel::configure(const ICLTensor *input1, const ICLTensor *input2, const ICLTensor *input3, const ICLTensor *input4, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input1, input2, input3, input4, output);
-}
-
-void CLWidthConcatenate4TensorsKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, const ICLTensor *input3, const ICLTensor *input4,
-                                                 ICLTensor *output)
+void CLWidthConcatenate4TensorsKernel::configure(const CLCompileContext &compile_context,
+                                                 ITensorInfo *input1, ITensorInfo *input2,
+                                                 ITensorInfo *input3, ITensorInfo *input4,
+                                                 ITensorInfo *output)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, input3, input4, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input1->info(), input2->info(), input3->info(), input4->info(), output->info()));
-
-    _input1 = input1;
-    _input2 = input2;
-    _input3 = input3;
-    _input4 = input4;
-    _output = output;
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input1, input2, input3, input4, output));
 
     // Add build options
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input1->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input1->data_type()));
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
-    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input1->info()->dimension(2)));
-    build_opts.add_option("-DINPUT1_WIDTH=" + support::cpp11::to_string(input1->info()->dimension(0)));
-    build_opts.add_option("-DINPUT2_WIDTH=" + support::cpp11::to_string(input2->info()->dimension(0)));
-    build_opts.add_option("-DINPUT3_WIDTH=" + support::cpp11::to_string(input3->info()->dimension(0)));
-    build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(input1->info()->element_size()));
+    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input1->dimension(2)));
+    build_opts.add_option("-DINPUT1_WIDTH=" + support::cpp11::to_string(input1->dimension(0)));
+    build_opts.add_option("-DINPUT2_WIDTH=" + support::cpp11::to_string(input2->dimension(0)));
+    build_opts.add_option("-DINPUT3_WIDTH=" + support::cpp11::to_string(input3->dimension(0)));
+    build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(input1->element_size()));
 
     // If input have different quantization info set quantization parameters needed for the re-quantization process
-    const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(output->info(), input1->info(), input2->info(), input3->info(), input4->info());
-    if(is_data_type_quantized_asymmetric(input1->info()->data_type()) && have_different_qinfo)
+    const bool have_different_qinfo = helpers::tensor_info::tensors_have_different_quantization_info(output, input1, input2, input3, input4);
+    if(is_data_type_quantized_asymmetric(input1->data_type()) && have_different_qinfo)
     {
-        const UniformQuantizationInfo iq1_info = input1->info()->quantization_info().uniform();
-        const UniformQuantizationInfo iq2_info = input2->info()->quantization_info().uniform();
-        const UniformQuantizationInfo iq3_info = input3->info()->quantization_info().uniform();
-        const UniformQuantizationInfo iq4_info = input4->info()->quantization_info().uniform();
-        const UniformQuantizationInfo oq_info  = output->info()->quantization_info().uniform();
+        const UniformQuantizationInfo iq1_info = input1->quantization_info().uniform();
+        const UniformQuantizationInfo iq2_info = input2->quantization_info().uniform();
+        const UniformQuantizationInfo iq3_info = input3->quantization_info().uniform();
+        const UniformQuantizationInfo iq4_info = input4->quantization_info().uniform();
+        const UniformQuantizationInfo oq_info  = output->quantization_info().uniform();
 
         build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq1_info.offset));
         build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
@@ -161,18 +152,18 @@
     _kernel = create_kernel(compile_context, "concatenate_width_x4", build_opts.options());
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input1->info(), input2->info(), input3->info(), input4->info(), output->info());
+    auto win_config = validate_and_configure_window(input1, input2, input3, input4, output);
     ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
 
     ICLKernel::configure_internal(std::get<1>(win_config));
 
     // Set output valid region
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
 
     // Pass paddings as arguments to the kernel
-    const unsigned int input1_width = input1->info()->dimension(0);
-    const unsigned int input2_width = input2->info()->dimension(0);
-    const unsigned int input3_width = input3->info()->dimension(0);
+    const unsigned int input1_width = input1->dimension(0);
+    const unsigned int input2_width = input2->dimension(0);
+    const unsigned int input3_width = input3->dimension(0);
 
     const unsigned int input1_right_padding = ceil_to_multiple(input1_width, num_elems_processed_per_iteration) - input1_width;
     const unsigned int input2_left_padding  = input1_width % num_elems_processed_per_iteration;
@@ -192,40 +183,47 @@
 
     // Set config_id for enabling LWS tuning
     _config_id = "concatenate_width_x4_";
-    _config_id += lower_string(string_from_data_type(input1->info()->data_type()));
+    _config_id += lower_string(string_from_data_type(input1->data_type()));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input1->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input1->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input1->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input1->dimension(1));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input2->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input2->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input2->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input2->dimension(1));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input3->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input3->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input3->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input3->dimension(1));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input4->info()->dimension(0));
+    _config_id += support::cpp11::to_string(input4->dimension(0));
     _config_id += "_";
-    _config_id += support::cpp11::to_string(input4->info()->dimension(1));
+    _config_id += support::cpp11::to_string(input4->dimension(1));
 }
 
-void CLWidthConcatenate4TensorsKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLWidthConcatenate4TensorsKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                                              const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC_VEC));
+    const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC_VEC + 1));
+    const auto src2 = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC_VEC + 2));
+    const auto src3 = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC_VEC + 3));
+    auto       dst  = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
     Window slice = window.first_slice_window_4D();
 
     do
     {
         unsigned int idx = 0;
-        add_4D_tensor_argument(idx, _input1, slice);
-        add_4D_tensor_argument(idx, _input2, slice);
-        add_4D_tensor_argument(idx, _input3, slice);
-        add_4D_tensor_argument(idx, _input4, slice);
-        add_4D_tensor_argument(idx, _output, slice);
+        add_4D_tensor_argument(idx, src0, slice);
+        add_4D_tensor_argument(idx, src1, slice);
+        add_4D_tensor_argument(idx, src2, slice);
+        add_4D_tensor_argument(idx, src3, slice);
+        add_4D_tensor_argument(idx, dst, slice);
         enqueue(queue, *this, window, lws_hint());
     }
     while(window.slide_window_slice_4D(slice));
diff --git a/src/core/CL/kernels/CLWidthConcatenateLayerKernel.cpp b/src/core/CL/kernels/CLWidthConcatenateLayerKernel.cpp
index 20cb962..591c26f 100644
--- a/src/core/CL/kernels/CLWidthConcatenateLayerKernel.cpp
+++ b/src/core/CL/kernels/CLWidthConcatenateLayerKernel.cpp
@@ -28,9 +28,9 @@
 #include "arm_compute/core/CL/CLValidate.h"
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/Window.h"
+#include "arm_compute/core/utils/misc/Cast.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 #include "support/StringSupport.h"
@@ -74,7 +74,7 @@
 } // namespace
 
 CLWidthConcatenateLayerKernel::CLWidthConcatenateLayerKernel()
-    : _input(nullptr), _output(nullptr), _width_offset(0)
+    : _width_offset(0)
 {
 }
 
@@ -85,31 +85,24 @@
     return Status{};
 }
 
-void CLWidthConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int width_offset, ICLTensor *output)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, width_offset, output);
-}
-
-void CLWidthConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int width_offset, ICLTensor *output)
+void CLWidthConcatenateLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, unsigned int width_offset, ITensorInfo *output)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), width_offset, output->info()));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, width_offset, output));
 
-    _input        = input;
-    _output       = output;
     _width_offset = width_offset;
 
     // Add build options
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->data_type()));
     build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
     build_opts.add_option("-DWIDTH_OFFSET=" + support::cpp11::to_string(_width_offset));
-    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->info()->dimension(2)));
+    build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->dimension(2)));
 
-    if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info())
+    if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info())
     {
-        const UniformQuantizationInfo iqinfo = input->info()->quantization_info().uniform();
-        const UniformQuantizationInfo oqinfo = output->info()->quantization_info().uniform();
+        const UniformQuantizationInfo iqinfo = input->quantization_info().uniform();
+        const UniformQuantizationInfo oqinfo = output->quantization_info().uniform();
 
         build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iqinfo.offset));
         build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oqinfo.offset));
@@ -120,23 +113,27 @@
     // Create kernel
     _kernel = create_kernel(compile_context, "concatenate_width", build_opts.options());
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), width_offset, output->info());
+    auto win_config = validate_and_configure_window(input, width_offset, output);
     ARM_COMPUTE_ERROR_THROW_ON(std::get<0>(win_config));
 
     ICLKernel::configure_internal(std::get<1>(win_config));
 
     // Set output valid region
-    output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
+    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
 }
 
-void CLWidthConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue)
+void CLWidthConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs,
+                                           const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(inputs.at(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(outputs.at(TensorType::ACL_DST));
+
     unsigned int idx = 0;
-    add_4D_tensor_argument(idx, _input, window);
-    add_4D_tensor_argument(idx, _output, window);
+    add_4D_tensor_argument(idx, src, window);
+    add_4D_tensor_argument(idx, dst, window);
     enqueue(queue, *this, window, lws_hint());
 }
 } // namespace arm_compute