We investigate in this article the use of reduction techniques in Air Pollution Modeling. The reduction of chemical kinetics is performed on the basis of a time-scale analysis and of lumping. Lumping techniques are widely used in Air Pollution Modeling and consist in replacing some {\it pure} chemical species by linear combinations of species. We focus here on the theoretical justification of such techniques. We propose an algorithm in order to build up lumped species in a systematic way. An application to three kinetic schemes coupled with diffusion is presented in a monodimensional case. This justifies the way we couple a reduced kinetic scheme with other processes.