An alternative to the procedure Erik described is to define two load steps in Details of Analysis Settings. You can turn time integration effects off for the first step, making it a steady state calculation. Apply boundary conditions that will give you the initial nonuniform temperature distribution that you seek.
Leave time integration effects on for the second step so that it becomes a transient that starts out with the non-uniform temperature calculated in step 1. You can activate/deactivate boundary conditions at the start of transient step 2 as needed to be consistent with the physical process you are simulating.