You should be aware of the fact that LES is not a turbulence model but a formulation that conceptually differs largely from RANS (where k-eps, k-om are used).

If your flow problem is at moderate characteristic lenght scale, if your computational resource allows you to compute a time-dependent and 3D problem, then I suggest to adopt LES using the dynamic model.

K-epsilon model

LES

It is not easy to determine the right turbulence model from beginning . You should have good experience to choose the suitable one that depends on the application you work with . However you should try more than one model and compare between the predicted results and to what level of accuracy these numerical results suit the experimental measurements.