The majority of control designs for wind power systems rest upon the simplifying assumption that every wind turbine and generator (WTG) inside a wind farm has identical dynamics, and, hence, can be aggregated into a single WTG admitting a single controller. In reality, however, this assumption does not hold because of wake effect due to which the small-signal models of turbines in a given row may be similar, but those between different rows can be significantly dissimilar. In this paper we propose a hierarchical control design to address these types of heterogeneous models of WTGs in a wind farm. Our approach is to group all turbines in a row as a cluster, design a cluster-level aggregate controller by averaging of states, make these aggregate controllers exchange average state information through a local communication network, and thereafter project the aggregate controller back to the original WTGs for implementation. Simulations validating the effectiveness of this hierarchical controller are shown for the IEEE 68-bus benchmark power system with one wind farm.