According to @_avichawla, kernels compute dot products in higher spaces without mapping data, enabling SVM and KernelPCA efficiency; RBF implies infinite dimensions.