Smartwatches have undergone rapid development. However, their small screen sizes limit their operational capabilities. The reason for this limitation lies in the fact that the small screen area is frequently obstructed by fingers during touch interactions. This study actively focuses on around-smartwatch gestures and introduces a new gesture recognizer model that uses a cost-effective built-in accelerometer to mitigate this challenge. To investigate the effectiveness of the proposed model, eight participants executed nine types of gestures involving arm motions and swiping on the skin in proximity to the smartwatch. The collected samples were employed to train an integrated recurrent convolutional neural network for developing a gesture recognition model. The model exhibited an accuracy of > 85%, demonstrating a promising use of gestures for effectively manipulating smartwatches.