diff --git a/src/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp b/src/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp
index 4a01c77..413c70a 100644
--- a/src/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp
+++ b/src/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.cpp
@@ -55,6 +55,7 @@
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(((lhs_info.k0 & (lhs_info.k0 - 1)) && lhs_info.k0 != 3), "Only 2,3,4,8,16 are supported for k0");
     ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.k0 > 16);
     ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.m0 < 2 || lhs_info.m0 > 8);
+    ARM_COMPUTE_RETURN_ERROR_ON((lhs_info.m0 > 4 && lhs_info.m0 < 8) && lhs_info.transpose);
 
     ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
     ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
@@ -70,11 +71,10 @@
     return Status{};
 }
 
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst, const GEMMLHSMatrixInfo &lhs_info, bool reinterpret_input_as_3d)
+Window configure_window(ITensorInfo *src, ITensorInfo *dst, const GEMMLHSMatrixInfo &lhs_info, bool reinterpret_input_as_3d)
 {
     const unsigned int num_elems_processed_per_iteration_x = lhs_info.k0;
     const unsigned int num_elems_processed_per_iteration_y = lhs_info.m0;
-    bool               window_changed                      = false;
 
     TensorInfo tmp_info(*src);
 
@@ -91,23 +91,13 @@
     auto_init_if_empty(*dst, src->clone()->set_tensor_shape(misc::shape_calculator::compute_lhs_reshaped_shape(*src, lhs_info, reinterpret_input_as_3d)));
 
     // Configure window
-    Window win    = calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-    Window win_in = calculate_max_window(*src, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-    AccessWindowStatic src_access(src, 0, 0,
-                                  src->dimension(0),
-                                  src->dimension(1));
-    AccessWindowStatic dst_access(dst, 0, 0, dst->dimension(0), dst->dimension(1));
-
-    window_changed = update_window_and_padding(win_in, src_access) || // window used by the execute_window_loop
-                     update_window_and_padding(win, dst_access);      // window used to update the padding requirements of dst tensor
+    Window win = calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
 
     // Collapse along the Z direction
     // This collapse needs to be here in order to tune the Z dimension of LWS
     Window collapsed = win.collapse(win, Window::DimZ);
 
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
-    return std::make_pair(err, collapsed);
+    return collapsed;
 }
 } // namespace
 
@@ -125,27 +115,20 @@
 
     auto padding_info = get_padding_info({ src });
 
-    _reinterpret_input_as_3d = reinterpret_input_as_3d;
-
-    const unsigned int src_w           = src->dimension(0);
-    const unsigned int src_h           = _reinterpret_input_as_3d ? src->dimension(1) * src->dimension(2) : src->dimension(1);
-    const unsigned int partial_load_m0 = src_h % lhs_info.m0;
-    const unsigned int partial_load_k0 = src_w % lhs_info.k0;
+    const unsigned int src_w      = src->dimension(0);
+    const unsigned int m          = reinterpret_input_as_3d ? src->dimension(1) * src->dimension(2) : src->dimension(1);
+    const unsigned int partial_m0 = m % lhs_info.m0;
+    const unsigned int partial_k0 = src_w % lhs_info.k0;
 
     // Create build options
     CLBuildOptions build_opts;
     build_opts.add_option("-DM0=" + support::cpp11::to_string(lhs_info.m0));
     build_opts.add_option("-DK0=" + support::cpp11::to_string(lhs_info.k0));
-    build_opts.add_option("-DV0=" + support::cpp11::to_string(lhs_info.v0));
-    build_opts.add_option("-DSRC_WIDTH=" + support::cpp11::to_string(src_w));
-    build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(src_h));
     build_opts.add_option_if(lhs_info.interleave, "-DINTERLEAVE");
-    build_opts.add_option_if(_reinterpret_input_as_3d, "-DREINTERPRET_INPUT_AS_3D");
-    build_opts.add_option_if(_reinterpret_input_as_3d, "-DHEIGHT_GEMM3D=" + support::cpp11::to_string(src->dimension(1)));
-    build_opts.add_option_if(_reinterpret_input_as_3d, "-DDEPTH_GEMM3D=" + support::cpp11::to_string(src->dimension(2)));
+    build_opts.add_option_if_else(lhs_info.transpose, "-DRESHAPE_LHS_T", "-DRESHAPE_LHS_NT");
     build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(src->element_size()));
-    build_opts.add_option("-DPARTIAL_LOAD_M0=" + support::cpp11::to_string(partial_load_m0));
-    build_opts.add_option("-DPARTIAL_LOAD_K0=" + support::cpp11::to_string(partial_load_k0));
+    build_opts.add_option("-DPARTIAL_M0=" + support::cpp11::to_string(partial_m0));
+    build_opts.add_option("-DPARTIAL_K0=" + support::cpp11::to_string(partial_k0));
 
     std::string kernel_name("gemm_reshape_lhs_matrix_");
     kernel_name += lhs_info.transpose ? "t" : "nt";
@@ -154,13 +137,16 @@
     _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(src, dst, lhs_info, reinterpret_input_as_3d);
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    ICLKernel::configure_internal(win_config.second);
+    auto win_config = configure_window(src, dst, lhs_info, reinterpret_input_as_3d);
+    ICLKernel::configure_internal(win_config);
+
+    unsigned int idx = 2 * num_arguments_per_3d_tensor_nhw();
+    _kernel.setArg<cl_int>(idx++, m);
+    _kernel.setArg<cl_int>(idx++, lhs_info.v0);
 
     // Set config_id for enabling LWS tuning
     _config_id = "gemm_reshape_lhs_matrix_";
-    _config_id += (_reinterpret_input_as_3d ? "3d_" : "");
+    _config_id += (reinterpret_input_as_3d ? "3d_" : "");
     _config_id += lower_string(string_from_data_type(src->data_type()));
     _config_id += "_";
     _config_id += support::cpp11::to_string(dst->dimension(0));
@@ -185,8 +171,6 @@
 Status ClGemmReshapeLhsMatrixKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const GEMMLHSMatrixInfo &lhs_info, bool reinterpret_input_as_3d)
 {
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, lhs_info, reinterpret_input_as_3d));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(src->clone().get(), dst->clone().get(), lhs_info, reinterpret_input_as_3d).first);
-
     return Status{};
 }
 
@@ -202,19 +186,11 @@
 
     Window slice = window.first_slice_window_3D();
 
-    if(_reinterpret_input_as_3d)
-    {
-        // Pass bottom paddings to the kernel if the src has to be reinterpreted as 3D tensor
-        const unsigned int idx0                  = 2 * num_arguments_per_3D_tensor();
-        const unsigned int total_cross_plane_pad = src->info()->padding().top + src->info()->padding().bottom;
-        _kernel.setArg<cl_uint>(idx0, static_cast<unsigned int>(total_cross_plane_pad));
-    }
-
     do
     {
         unsigned int idx = 0;
-        add_3D_tensor_argument(idx, src, slice);
-        add_3D_tensor_argument(idx, dst, slice);
+        add_3d_tensor_nhw_argument(idx, src);
+        add_3d_tensor_nhw_argument(idx, dst);
         enqueue(queue, *this, slice, lws_hint());
     }
     while(window.slide_window_slice_3D(slice));