Deep learning creates complicated hierarchical models while training and the models executed during inference. To conduct out operations or training, these sophisticated models must be loaded onto a computer (hard disc, RAM, and GPU) due to which certain complications arises in terms of working speed of a computer and time taken by the training operations. This paper shows how to compute the amount of space necessary for such training operations for a fully connected networks and deep convolutional networks that is, model space, operational space & operational compute complexities along with the computation of batch size which helps in proper and appropriate selection of computer architecture for AI systems.