Customizing computer vision applications for embedded systems is a common and widespread problem in the cyber-physical systems (CPS) community. Such a customization has two main goals: (i) parametrizing the application algorithm by considering the external environment, and (ii) correctly mapping the Software application to the heterogeneous processing components of the target Hardware to satisfy non-functional constraints like performance, power, and energy consumption. The lack of automation and verification methodologies in such design ow makes embedded software development a time consuming and error prone task especially in real and complex contexts. One of them is ORB-SLAM, a novel application that computes in real-time the image sensor trajectory and a sparse 3D reconstruction of the scene taken from an agent moving around indoor or outdoor environments. It is a key application for autonomous navigation in the Robotics field. What is missing is a model-based approach for the design, parametrization, and verification of embedded vision applications targeting CPS. This project aims at defining such an approach and to apply it to the ORB-SLAM case study. This gives two fundamental contributions to the CPS community: a general and efficient methodology for the design, customization, and verification of embedded applications and a reference implementation of ORB-SLAM ready to be synthesised into different existing low-power embedded systems.