Artificial intelligence (AI) applications in edge computing, especially computer vision, are expanding quickly, necessitating the need for systems that can effectively and instantly implement deep learning models. FPGA architectures offer an appropriate platform in this regard because of their reconfigurability, parallel processing, and high speed. However, most existing approaches address only parts of the design flow, and comprehensive end-to-end frameworks are rarely introduced. This work provides a thorough approach to implementing AI models on Zynq UltraScale+ FPGAs using Vitis AI and the Deep Processing Unit (DPU). Thanks to the high flexibility that the proposed framework offers in the choice of inputs and outputs, any desired model, including pre-trained models, can be used. Cameras or sensors can provide the inputs, and HDMI, LCD, or other interfaces can display the outputs. The steps in the implementation process include designing Vivado hardware with DPU integration, creating XSA files, building Petalinux images using official Vitis AI recipes, and executing quantized models on a board. The efficiency and versatility of the proposed framework are demonstrated by the Zynq UltraScale+ board's processing time per image using the MNIST dataset, which is approximately 0.21 ms (≈4,740 FPS).