Chemical flame structures encountered in practical turbulent combustion chambers are complex because multiple regimes such as premixed, stratified or diffusion may coexist. Combustion modeling strategies based on single flame archetype fail to predict pollutant species, such as CO. To account for multiple combustion regimes, at a reduced computational cost, a novel approach based on virtual optimized mechanisms is developed. This method consists in (i) building a kinetic scheme from scratch instead of reducing a detailed mechanism, (ii) using a reduced number of virtual reactions and virtual species that do not represent real entities and (iii) using sub-mechanisms dedicated to the prediction of given flame quantities. In the present work, kinetic rate parameters of the virtual reactions and virtual species thermody-namic properties are optimized through a genetic algorithm to properly capture the flame temperature as well as CO formation in hydrocarbon/air flames. The virtual optimized chemistry approach is first applied to the derivation of methane/air reduced kinetic schemes. The flame solutions obtained with the virtual optimized mechanisms are subsequently compared to the reference flame library showing good predictive capabilities for both premixed and non-premixed flame archetypes. Analysis of the impact of the reference database demonstrated that the quality of CO formation modeling depends on the reference flame library used to train the optimized kinetic scheme. The virtual mechanisms are also tested on 2-D partially-premixed burners showing good agreement between the reference mechanism and the reduced virtual schemes. Finally, the virtual optimized chemistry approach is used to derive virtual optimized mechanisms for heavy fuels oxidation.