How similar are two images? In computational pathology, where Whole Slide Images (WSIs) of digitally scanned tissue samples from patients can be multi-gigapixels in size, determination of degree of similarity between two WSIs is a challenging task with a number of practical applications. In this work, we explore a novel strategy based on kernelized Maximum Mean Discrepancy (MMD) analysis for determination of pairwise similarity between WSIs. The proposed approach works by calculating MMD between two WSIs using kernels over deep features of image patches. This allows representation of an entire dataset of WSIs as a kernel matrix for WSI level clustering, weakly-supervised prediction of TP-53 mutation status in breast cancer patients from their routine WSIs as well as survival analysis with state of the art prediction performance. We believe that this work will open up further avenues for application of WSI-level kernels for predictive and prognostic tasks in computational pathology.