This thesis takes place within the framework of statisticallearning. It brings contributions to the machine learning community usingmodern statistical techniques based on progress in the study of empiricalprocesses. The first part investigates the statistical properties of KernelPrincipal Component Analysis (KPCA). The behavior of the reconstruction error isstudied with a non-asymptotic point of view and concentration inequalities ofthe eigenvalues of the kernel matrix are provided. All these results correspond to fast convergence rates. Non-asymptotic results concerning theeigenspaces of KPCA themselves are also provided. A new algorithm ofclassification has been designed in the second part: the Kernel ProjectionMachine (KPM). It is inspir...