In this paper the client specific kernel discriminant analysis (CSKDA) is studied. The theory of CSKDA, which is the nonlinear model of the previously suggested model of client specific linear discriminant analysis, is proposed by using kernel technique. A new CSKDA subspace method is developed in order to reduce the computational complexity. Results of experiments conducted on the internationally recognized facial database of XM2VTS based on the Lausanne protocol show the effectiveness of the proposed methods of client specific kernel discriminant analysis.