The advantage of using SAS PROC KPCA is that you can preprocess your data so that you can classify groups with nonlinear classification boundaries.
SAS® Fast-KPCA implementation bypasses the limitations of exact KPCA methods. SAS® internally uses k-means to find a representative sample of a subset of points. This row reduction method has the advantage that c centroids are chosen to minimize the variation of points nearest to each centroid and maximize the variation to the other cluster centroids. In some cases, the downstream effect of using k-means on computing the SVD increases numerical stability and improves clustering, discrimination, and classification.