First determine which parts of your model are intended for embedded system deployment.
Isolate that portion into a subsystem or it's own reference model. Let's call that the System Under Design (SUD).
Next determine appropriate discrete time sample rates, then use Model Discretizer to convert the SUD from continuous-time to discrete-time.
A key aspect of fixed-point is not wasting resources covering signal ranges that will never occur.
For this goal, determine the ranges of the key inputs to the SUD.
For example, suppose one of the inputs to the SUD is ambient air temperature. For a device that will operate in rugged outdoor environments, the range is something like -40F to 130F. The fixed-point design only needs to handle this range.
Another key aspect of fixed-point is not wasting resources providing precision levels that are needlessly luxurious.
To support the goal of setting wise precision levels, one approach is to have good simulation scenarios to measure the behavioral performance of the system over a rich set of operating points.
Using a unit step response as is done in Control Systems text books is NOT adequate. That approach totally depends on the system being linear. The assumption of linearity does not hold when using fixed-point to minimize resource consumption. As an operating point gets closer to the range or precision limits of a fixed-point design, the behavior will become more non-linear.
By simulating the design over a rich set of operating scenarios the fixed-point design can be properly tuned to the needs of the real world embedded system.
Once this preparation is done, use the Fixed-Point Tool or fxpopt to convert the SUD portion of the model to fixed-point.
