Text this: Micro-kernels for portable and efficient matrix multiplication in deep learning