Optimization of a CNN-to-Vector Processor Conversion Tool for State-of-the-Art Convolutional Neural Network Architectures