This study proposes adaptive downlink association schemes for controlled device-to-device (D2D) communication in cellular networks. The proposed schemes utilise active devices under specific conditions to improve the network performance. Specifically, active devices are classified into two disjoint groups according to their individual quality of service (QoS) requirements from their base station (BS). The BS adaptively allocates downlink physical channels to meet individual QoS of devices using a minimal number of these available channels. The unused channels at each served device (in the first class) can be then utilised by that device to serve other devices from the second class, which are not served by the BS, via controlled D2D associations. Herein, D2D pair discovery as well as the conditions for establishing successful D2D association between the classified devices are treated. Furthermore, two D2D association schemes that vary in terms of their performance and implementation complexity to meet certain objectives at the device of interest are presented. The developed analytical results address the scenarios of idealised perfect and practical imperfect D2D association. Numerical results are provided to further explain the performance variations between the proposed association schemes under perfect and imperfect operation scenarios.