Text this: Optimization of Direct Convolution Algorithms on ARM Processors for Deep Learning Inference