How to use the proxy for everything but Google
If you are hijacking the traffic to send it through a "transparent" proxy, then you could configure the device doing the hijacking to not hijack the IP ranges known to belong to Google. Then those requests won't go through the proxy.
If you are rather relying on a web proxy auto detection script, then you can update the script to consider the hostname before deciding on whether to use a proxy or direct connection.
However if those requests end up all going through a single NAT instead of all through a single proxy, Google may never know the difference. It would still see all the same requests coming from a single IP, which may look equally suspicious if done through NAT or proxy.
It may be the only way you could get a separate IP for each device is by using IPv6. Luckily Google supports IPv6, so if you do get IPv6 on your network, Google will be able to tell the difference between requests from different devices.
There may be a single or a few devices on your network, which are flooding Google with abusive requests. Try to track down the devices doing the most requests to Google, and figure out if they are legitimate. If you have a few machines infected by botnets, then cleaning them might solve the problem.
You could also double check if your proxy is sending all information to Google, which could help Google identify clients. If Google can tell which requests came from which device, they might only block the abuse devices and not everything going through your proxy. For
http you could ensure all requests to Google get an
X-Forwarded-For header. It might be Google would entirely ignore that header, you can really only find out by trying.
If the requests are done over
https there is not much you can do. You can't help Google identify individual clients, and you can't figure out what requests are performed, only the volume. However the volume of requests might be enough to identify if there are any abusive devices on your network.