In this paper we define distance functions for data sets (and distributions) in a RKHS context. To this aim we introduce kernels for data sets that provide a metrization of the set of points sets (the power set). An interesting point in the proposed kernel distance is that it takes into account...