Near-wall turbulence modeling is computationally a very expensive problem. The talk considers a novel approach based on non-overlapping domain decomposition. It allows us to avoid calculations of the region with high gradients in the vicinity of the wall while retaining sufficient overall accuracy. The technique is introduced in application to low-Reynolds number RANS models. The domain decomposition is achieved via the transfer of the boundary condition from the wall to an interface boundary. If the governing equations in the inner domain are simplified, then the interface boundary conditions are of Robin type. These boundary conditions can be obtained in an analytical form despite the fact that they are nonlinear. Possible ways to achieve a reasonable trade-off between efficiency and accuracy are discussed. The obtained interface boundary conditions are mesh-independent. They can be used to avoid the computationally expensive resolution of a high-gradient region near the wall. Moreover, once the solution is constructed in the outer region, the near-wall profile can be restored if required. In two extreme cases, if the interface boundary is too close to the wall or too far from it, the so-constructed solution to the problem automatically corresponds to low- and high-Reynolds number RANS models, respectively. Different applications are considered including unsteady problems and complex geometries. The developed approach proved to be quite robust and relatively universal. It does not contain any tuning parameters. The technique might be extended to other multiscale problems.