In most practical applications, a large variety of flame structures varying from premixed, to non-premixed combustion is encountered. To account for such complex combustion regimes, at a reduced computational cost, predictive reduced kinetic schemes are required. Global mechanisms are widely used in the LES context for their low computational cost and easy implementation in CFD codes. However, these mechanisms, built to reproduce global quantities such as laminar flame speed and burnt gas state, are valid on a limited range of flame equivalence ratio, and do not reproduce the local flame structure. A strategy to optimize virtual kinetic schemes based on a genetic algorithm is discussed. The optimization procedure aims at evaluating the set of virtual species properties and reaction rate parameters that best reproduce the temperature profile and laminar flame speed of a reference library of premixed flames. The use of virtual species whose thermodynamic properties are fitted enable the prediction of adiabatic temperature over the whole flammability range. The model also uses a correction for pre-exponential factors to match the laminar flame speed in very lean and rich flames. Furthermore, it is proposed to tabulate the reaction exponent of an intermediate species versus the equivalence ratio in order to adjust the characteristic thickness of the post-flame zone. The resulting kinetic scheme is evaluated on premixed and non-premixed flame configurations. The temperature profiles predicted by the virtual mechanism are in good agreement with these obtained from detailed chemistry calculations for both combustion regimes.