Are you planning on modeling the model ball explicitly (meaning having the outer surfaces of the round balls meshed)? This approach may be computationally challenging if the diameter of your cylinder is orders of magnitude larger than that of the balls. Also the fill height of the balls is an important consideration.

Or do you just want to draw a cylinder and cut out where the packed bed of balls would be? You would then represent this region as a porous medium in Fluent. The porous media properties in the domain will be set to approximate their effects the flow field (for example pressure drop). These porous media properties could come from experiment or correlations you found in the literature.

The third option is we would create two separate models. We could create a model of a representative block of balls. In this one we would resolve the curvatures of the balls. But instead of creating the full bed, we would only create a small fraction of the full cylinder. This could be simulated to give the properties for the porous media of the larger model and avoid having to obtain experimental measurements.

How one would create the above two models in SpaceClaim or DesignModeler would be different. So we would need to know your approach before we could provide a recommendation.