Demand responsive nodepool scaling in Kubernetes

As a customer of managed Kubernetes, I wish for demand responsive autoscaling of node pools below 1 node per nodepool. 

Use case: I want to deploy CI/CD runners in the cluster. The runners will need large machines (64 GB memory) and will only run irregularly (e.g. once a week for 12 hours). Hence, it would make sense to only spawn one of these machines if they are REALLY needed.

The need should be determined by one main factor and one side factor:

1. The label of the nodepool (e.g. purpose=xlarge.runner)

2. The requested resources (side factor/optional)

The node pool should be at zero nodes (not 1) per default and scale down after no resource with the given label and resource need exists.

I think this feature would benefit other customers as well and would be a good feature for a blog post or similar, as not many managed Kubernetes providers have this. :)

Best regards