Worked it out using a previous thread, doing it using an affine transform on the window. 
I would however like to know if the rasterio.transform.xy method can be used when passing col, row from a dataset window, as I would like to get the coordinate for the center of the cell rather than the ul which results from the affine window transform.

Does a window generated by rasterio maintain the dataset origin and absolute col/row indices?

I'd like to be able to iterate over a window of a dem dataset, find the highest value in the window and return it's lat long coordinates using rasterio.transform.xy or similar.

Obviously the below isn't suitable (the arr is invalid), it illustrates what I'm trying to achieve. Window is generated using slices and row/col high and lows which are provided by a valid function called extract_block.

with rasterio.open(inputraster) as dataset:
        for (x,y) in zip(*nonzero):
            row_low, row_high, col_low, col_high = extract_block(arr,[x,y],windowsize)
            rasteriowindowarr = dataset.read(1,window=Window.from_slices(slice(row_low, row_high), slice(col_low, col_high)))
            with rasterio.open(rasteriowindowarr) as windowarr:
                for i,j in np.ndindex(rasteriowindowarr.shape):
                    coordinatex,coordinatey = rasterio.transform.xy(windowarr.transform, i,j,offset='center')
                    print ("long of result is " + str(coordinatex))
                    print ("lat of result is " + str(coordinatey))