Air brake operations can generate large values of compressive in-train forces, which can eventually be related to an increase of the derailment risk. Longitudinal train dynamics (LTD) simulations are commonly run to compute the in-train forces. However, typical LTD codes are not able to calculate the wheel-rail forces, which are needed for the evaluation of the safety indexes defined by the international standards. On the other hand, wheel-rail contact forces can be easily determined by means of multibody (MB) codes, but MB simulations of whole trains including many wagons are usually too expensive from a computational point of view. The present paper proposes to obtain metamodels of a single wagon built via kernel-based regressions, trained from the results of LTD and MB simulations of freight train air brake operations. Typical wagons running in Europe are considered as reference vehicles in the simulations. The derived metamodels are closed-form models that can be included in common LTD simulators for a fast evaluation of the safety indexes directly from the main outputs of LTD simulations of braking operations.