HB estimation is similar to Latent Class in that you use the original choice data and come up with a model. However, instead of group-level models, HB will come up with a model for each individual respondent.
You run HB very similarly to Latent Class. If you are on older versions of the software, export out your .cho file and use the standalone CBC/HB software. If you are on a more current version, select HB from the analysis menu.
Most people would use the HB utilities to calculate share predictions in a simulator, and use the Latent Class group assignments as segmentation variables.