Deep learning models for food recognition that are now available do not allow for data incremental learning and frequently suffer from catastrophic interference difficulties during class incremental learning. Because real-world food databases are open-ended and dynamic, with a constant rise in food samples and food classes, this is a critical challenge in food recognition. To deal with the dynamic nature of the data, model retraining is frequently used, although it necessitates high-end computer resources and a significant amount of time. By combining transfer learning on deep models for feature extraction, Relief F for feature selection, and an unique adaptive reduced class incremental kernel extreme learning machine (ARCIKELM) for classification, this study offers a new open-ended continual learning framework. Transfer learning is advantageous because deep learning has a strong generalisation capacity.