Short Description: The exponential growth of Deep Learning (DL) in recent years led to a plethora of different frameworks for training and deploying DL models. However, most frameworks use their own formats and conventions for defining, storing, and sharing data, models, and metadata. This fragmented landscape has negatively impacted DL-related research and development, since any tool, optimizer, and hardware platform has to support several different data formats and protocols, leading to increased costs and complexity. This has fueled the interest in developing interoperable standards in DL, such as Open Neural Network Exchange (ONNX). In this talk, we will briefly explore the DL framework landscape, focusing on standards, data formats, and ways to ensure interoperability between different frameworks and tools. Then, we will outline the technical solutions, regarding both the architecture and data formats, currently adopted by OpenDR to develop an agile, interoperable, yet practical and easy to use framework for supporting deep learning-enabled robotics applications.