WRF is capable of running both real and ideal LES cases; however, if you plan to run the ideal case, you will need to compile the ideal LES case, select periodic BC and use single profiles of wind, temperature, mixing ratio, or dewpoint temperature/RH (please check), etc. as input. For a real LES case, you need to be cautious about case selection. It is computationally intensive, as your time scales are very small (may be a second or less, depending on the resolution you run). To have some realistic turbulence, you need to go down to a very high resolution, like 30-60 m. When you run LES, you are explicitly resolving eddies that carry most energy (i.e., you resolve a part of the turbulence spectrum which otherwise is not resolved in the mesoscale model). Some of the PBL schemes are derived from the LES studies. It is a good idea to use YSU in the outer domain when you run LES in the inner domain. In the upper levels, you need to have a maximum of 500m resolution, which means many vertical layers, and increasing the computation time. You could use PBL in coarser domains and LES in the finer ones.