Accelerating a target application is an important part of system design. For a given application, system designers may use several acceleration methods to speed-up critical code segments. These acceleration methods can be at software level (software optimization) or at hardware level using high-performance computing units, like multiprocessor platforms or ASICs. However, the resulting performances may vary from application to application because of their intrinsic characteristics. Among these applications, we focus on a 3-D vision application that reconstructs the 3D shape of a target object from data collected from the scene. This reconstruction is based on specific functions that can be implemented in software on a DSP, or in hardware using look-up tables (LUTs). In this paper, we introduce a new mathematical model that predicts, at system level, the acceleration that can be achieved when using LUTs implemented with different memory technologies. Simulated performance results will be shown for the reconstruction of a 3-D image with an autosynchronized optical camera. © 2003 IEEE.