<p>Edge computing faces critical bottlenecks in power efficiency and latency due to the physical limitations of electronic architectures. While optical computing presents a promising alternative, conventional coherent systems face significant challenges in edge deployment due to their sensitivity to environmental perturbations and complex laser requirements, which are particularly mismatched with the incoherent natural light conditions typical of edge environments. Freeform optical surfaces can encode incoherent spatial light information through Point Spread Function (PSF) method, showing great potential for the development of intelligent sensing lenses for edge devices. This paper comprehensively reviews the research history of freeform optics and extends to emerging design paradigms, particularly differentiable ray tracing methods and end-to-end software-hardware co-design approaches that enable gradient-based optimization of optical systems through seamless integration with neural networks. However, once fabricated, the encoding capabilities of freeform optical surfaces are fixed, making it difficult to support the reconfiguration of intelligent recognition functions. Furthermore, freeform optics lack the nonlinear recognition capabilities required for complex scenes. Therefore, this paper further reviews relevant frontier developments, including programmable optical modulators based on phase-change materials and optical nonlinear activation elements based on special materials and structures.</p>