Evaluating Pretrained models for Deployable Lifelong Learning