Helper methods for Conv2D kernel

- 16x8 helpers, one for 32-bit biases and one for 64-bit biases
- 8x8 helper (using aconv)

Change-Id: Ife5013ac8eceb51bbc4ca1091f2b465e5b315faf
diff --git a/tflm/opt/BUILD b/tflm/opt/BUILD
index a6722df..28419ce 100644
--- a/tflm/opt/BUILD
+++ b/tflm/opt/BUILD
@@ -3,6 +3,7 @@
 cc_library(
     name = "opt",
     srcs = [
+        "conv.cc",
         "elementwise_add_s16.cc",
         "elementwise_add_s32.cc",
         "elementwise_add_s8.cc",
@@ -17,6 +18,7 @@
     target_compatible_with = ["@kelvin_sw//platforms/cpu:kelvin"],
     deps = [
         "//crt:crt",
+        "@tflite-micro//tensorflow/lite/kernels/internal:common",
     ],
     alwayslink = True,
 )