In this study, we explore context-aware cross-device interactions between a smartphone and smartwatch. We present 24 contexts, and then examine and prioritize suitable user interfaces (UIs) for each. In addition, we present example applications, including a map, notification management system, multitasking application, music player, and video chat application, each of which has its own context-aware UIs. To support these context-aware UIs, we investigate the performance of our context recognizer in which recognition is based on machine-learning using the accelerometers in a smartphone and smartwatch. We conduct seven different evaluations using four machine-learning algorithms: J48 decision tree, sequential minimal optimization (SMO)-based support vector machine (SVM), random forest, and multilayer perceptron. With each algorithm, we conduct a long-interval experiment to examine the level of accuracy at which each context is recognized using data previously collected for training. The results show that SMO-based SVM is suitable for recognizing the 24 contexts considered in this study.