During the last few decades, air pollution has become one of the major environmental and public health issue in every important cities over the world. Photochemical pollutants, like ozone which play a central role in today's air pollution problems, are formed in the atmosphere by reaction of two emitted precursors: Volatile Organic Carbons (VOCs) and NOx (NO + NO2). Ozone is a highly non linear process because its formation is driven by complex chain reactions. The decrease of ozone concentration produced by a reduction of its emitted precursors is therefore unpredictable, unless calculated by numerical photochemical models. The simulation of photochemical air pollution requires detailed chemical mechanisms and a lot computer resources. The number of chemical species within the chemical mechanism has to be confined to the strict minimum in order to minimise the CPU time. A solution is to lump the immense number of VOC species involved in atmospheric pollution in a convenient smaller number of mechanism species, keeping enough details to generate accurate results in reasonable calculations times. The calculation of all kinetic data of a lumped mechanism is a tremendous work unless carried out by a generation programme. CHEMATA, presented in this work, is a chemical mechanisms generation programme designed to create lumped and explicit tropospheric gas phase chemical mechanisms. Based on the widely used mechanism RACM, CHEMATA generated an extended mechanism to test the carbonyl species parameterisation of RACM and two smaller mechanisms to compare two lumping methods (the reduced mechanism and the small mechanism). The new mechanisms have been implemented in a bOx model and in the 3D eulerian air quality model TAPOM, also presented in this work. TAPOM have been run with the four chemical mechanisms on three simulations domains (Mexico City, Milan and Bogota) presenting different emissions strengths and meteorological conditions. The comparisons between the different mechanisms, in a bOx or 3D model and with or without emission reductions lead to the following conclusions: The comparison between the extended mechanism and RACM shows that the treatment of the carbonyl species in RACM does not induce notable errors in mesoscale modelling. The use of the extended mechanism should be kept for special simulations when enhanced precision in VOCs is required or for time periods longer than 2 or 3 days. The reduced mechanism is the best compromise between CPU time and accuracy. When calculating photochemical pollution, or emission reduction scenarios, this mechanism can save a lot of time. The small mechanism presents a clear tendency to produce more "VOC sensitive" results, which can lead to severe ozone overestimations. It should only be used for qualitative simulation when CPU time is a critical issue.