Model theory, machine learning, and combinatorics each have generalizations of VC-dimension for fuzzy and real-valued versions of set systems. These different dimensions define a unique notion of a VC-class for both fuzzy sets and real-valued functions. We study these VC-classes, obtaining generalizations of certain combinatorial results from the discrete case. These include appropriate generalizations of $\varepsilon$-nets, the fractional Helly property and the $(p,q)$-theorem. We then apply these results to continuous logic. We prove that NIP for metric structures is equivalent to an appropriate generalization of honest definitions, which we use to study externally definable predicates and the Shelah expansion. We then examine distal metric structures, providing several equivalent characterizations, in terms of indiscernible sequences, distal types, strong honest definitions, and distal cell decompositions.